1 00:00:01,040 --> 00:00:05,484 You're about to join Niels Kaastrup-Larsen on a raw and honest 2 00:00:05,484 --> 00:00:09,524 journey into the world of systematic investing and learn about 3 00:00:09,524 --> 00:00:13,160 the most dependable and consistent yet often overlooked investment 4 00:00:13,160 --> 00:00:15,988 strategy. Welcome to the Systematic Investor Series. 5 00:00:23,360 --> 00:00:27,958 Welcome and welcome back to this week's edition of the Systematic 6 00:00:27,958 --> 00:00:31,720 Investor Series with Rob Carver and I, Niels Kaastrup-Larsen, 7 00:00:31,720 --> 00:00:36,736 where each week we take the pulse of the global market through 8 00:00:36,736 --> 00:00:39,244 the lens of a rules-based investor. Rob,it is great to be 9 00:00:27,500 --> 00:00:35,780 back with you this week. Hope you're doing well. How are things 10 00:00:35,780 --> 00:00:39,230 in the in the UK? 11 00:00:40,530 --> 00:00:43,914 It's cold and sunny. I think we're recording video for this episode, 12 00:00:43,914 --> 00:00:47,580 so, people can probably see I'm wearing a lot of layers as my 13 00:00:47,580 --> 00:00:49,836 garden office is not particularly well insulated. So, 14 00:00:49,836 --> 00:00:53,502 I've got a heater running, and I'm wearing lots of layers, and hopefully 15 00:00:53,502 --> 00:00:56,886 I won't freeze to the death before the end of the podcast. 16 00:00:57,250 --> 00:01:00,930 Now I'll make sure we'll keep you warm with some pointed topics 17 00:01:01,330 --> 00:01:01,850 for sure. 18 00:01:01,850 --> 00:01:04,010 If I get, if I get nice and angry, that will definitely keep 19 00:01:04,010 --> 00:01:04,530 me warm. 20 00:01:04,930 --> 00:01:09,830 Absolutely. And speaking of those, we've actually got quite a 21 00:01:09,830 --> 00:01:15,710 few questions this week, so that's going to keep you warm, I'm 22 00:01:15,710 --> 00:01:22,080 sure. And, as well as a very, very important topic that I personally 23 00:01:22,080 --> 00:01:27,960 think could be a game changer, not just for trend following and 24 00:01:27,960 --> 00:01:33,840 CTAs, but generally speaking. But we'll come to that a little bit 25 00:01:33,840 --> 00:01:34,330 later. Ofcourse, before we get to any of that, I'm always curious 26 00:01:16,702 --> 00:01:27,493 what you've been thinking about the last few weeks, since we 27 00:01:27,493 --> 00:01:34,360 spoke. Anything that stood out to you? 28 00:01:35,410 --> 00:01:38,098 Well, as always, I've been writing on my blog and people can 29 00:01:38,098 --> 00:01:41,010 go and look there, but I'm not here to plug my blog, so. 30 00:01:41,410 --> 00:01:42,050 Of course not. 31 00:01:42,050 --> 00:03:38,126 There are some very good articles on there, I have to say. 32 00:01:42,050 --> 00:02:03,806 Actually,the thing that kind of came to my head this morning, 33 00:02:03,806 --> 00:02:29,188 I was trying to think of stuff to talk about was actually a discussion 34 00:02:29,188 --> 00:02:54,570 I had on Twitter (now known as X) yesterday. And it's quite an interesting 35 00:02:54,570 --> 00:03:14,513 discussion, it comes up every now and then, and it's basically 36 00:03:14,513 --> 00:03:36,269 this discussion about whether trend following is an edge or a risk 37 00:03:36,269 --> 00:03:38,082 premium. Andof course, you know, I think the word ‘edge’ is 38 00:02:09,209 --> 00:02:38,837 kind of one of these words that people use in different ways. 39 00:02:38,837 --> 00:03:13,403 So, I think the most loose sense of it is basically if you can 40 00:03:13,403 --> 00:03:38,093 make money in the markets and you have an edge. Mypersonal belief 41 00:01:45,713 --> 00:02:02,807 is, if anyone can do it then you haven't got an edge (If you're 42 00:02:02,807 --> 00:02:16,238 making money out of something, but it's something that anyone can 43 00:02:16,238 --> 00:02:30,890 do). And, of course, anyone can sort of follow a simple trend 44 00:02:30,890 --> 00:02:43,100 following system if they've got the right amount of capital, 45 00:02:43,100 --> 00:02:56,531 the right resources and make money, hopefully, in the long run. 46 00:02:56,531 --> 00:03:14,846 Then that money must be coming from what I would see as a risk premium. 47 00:03:14,846 --> 00:03:25,835 So, basically you're only making money because other people 48 00:03:25,835 --> 00:03:38,045 are uncomfortable taking the kinds of risks that you're taking. 49 00:01:42,050 --> 00:02:04,370 So,we can talk about psychological biases, we can talk 50 00:02:04,370 --> 00:02:26,690 about delayed reactions to information. Those are the main two 51 00:02:26,690 --> 00:02:49,010 explanations for why trend following makes money. In fact, there's 52 00:02:49,010 --> 00:03:13,562 a recent Odd Lots podcast, with Cliff Asness, where he talked 53 00:03:13,562 --> 00:03:38,114 about that exact subject. So, that's worth looking up as well. 54 00:01:42,050 --> 00:01:57,065 Soyeah, it was kind of interesting because this debate comes 55 00:01:57,065 --> 00:02:13,445 along every now and then and it's amazing how heated people get 56 00:02:13,445 --> 00:02:28,460 about this stuff. So, very insistent that, oh no, there's definitely 57 00:02:28,460 --> 00:02:46,205 a trend here, an edge in using, I don't know, a moving average 58 00:02:46,205 --> 00:03:03,950 crossover or something to pick up trends. How could it be an edge 59 00:03:03,950 --> 00:03:20,330 when literally anyone with a spreadsheet can do it or even just 60 00:03:20,330 --> 00:03:38,075 a calculator? So, that was the thing that got me heated up yesterday. 61 00:03:38,130 --> 00:03:47,980 Yeah, it's kind of interesting. I mean, I think sometimes 62 00:03:47,980 --> 00:04:00,785 we also try and overcomplicate stuff that we have to kind of be 63 00:04:00,785 --> 00:04:12,605 able to explain in a very academic way. So, yeah, I'll check 64 00:04:12,605 --> 00:04:24,425 it out. And by the way, the conversation with Cliff Asness, it's 65 00:04:24,425 --> 00:04:38,215 queued up on my phone because every time he speaks I'm sure I learn 66 00:04:38,215 --> 00:04:52,005 something and have a laugh at the same time. So, he's one of my 67 00:04:52,005 --> 00:05:03,825 favorites to listen to. So, I'll definitely make sure I get that. 68 00:03:38,130 --> 00:05:03,830 Acoupleof things that stood out on what's been on my radar… Don't 69 00:03:46,628 --> 00:03:55,733 know if you saw this. It kind of helps make CTAs not look so bad 70 00:03:55,733 --> 00:04:02,410 this year. And that is what happened over with the brilliant 71 00:04:02,410 --> 00:04:07,873 people at Renaissance Technologies, in October. I did notice 72 00:04:07,873 --> 00:04:15,157 that, in an article, a couple of their funds were behaving very 73 00:04:15,157 --> 00:04:20,013 differently. The Renaissance Institutional Equity Fund was down 74 00:04:20,013 --> 00:04:27,904 14% in October and down 8.3% so far this year. And the diversified 75 00:04:27,904 --> 00:04:34,581 Alpha strategy lost more than 15% last month, according to the 76 00:04:34,581 --> 00:04:41,258 article, putting it down more than 10.5% last month. Of course, 77 00:04:41,258 --> 00:04:48,542 what they're really known for is none of that, it's The Medallion 78 00:04:48,542 --> 00:04:55,219 fund, which they didn't quote the performance. I have a sneaky 79 00:04:55,219 --> 00:05:01,896 feeling it might have made money last month, but that's purely 80 00:05:01,896 --> 00:05:03,830 on my side. Yeah,interesting. 81 00:05:04,550 --> 00:05:12,899 Yeah. I mean there's always been a big disparity between the 82 00:05:12,899 --> 00:05:21,248 Medallion fund, which has been closed forever. I think it's only 83 00:05:21,248 --> 00:05:29,597 a very few small number of non-outsiders that have money in 84 00:05:29,597 --> 00:05:40,223 it. Most of the money in there belongs to the employees or the beneficiaries 85 00:05:40,223 --> 00:05:49,331 of the wills of the employees. Because, of course, Jim Simons is 86 00:05:49,331 --> 00:05:52,367 no longer with us. Soyeah, there's always been a big disparity 87 00:05:09,654 --> 00:05:16,034 between the returns of Medallion, which is this highly secretive 88 00:05:16,034 --> 00:05:23,690 fund. No one really knows how much money it makes either, because 89 00:05:23,690 --> 00:05:30,708 there's no public information out there. And the returns are the 90 00:05:30,708 --> 00:05:39,002 funds that you and I can buy that Renaissance sells. And I remember 91 00:05:39,002 --> 00:05:46,658 them, they’re launching a CTA type strategy. I think it was called 92 00:05:46,658 --> 00:05:52,400 the Renaissance Institutional Futures Fund or something like that. 93 00:05:52,650 --> 00:05:53,210 Okay. 94 00:05:53,290 --> 00:06:03,298 I remember I had a catch acronym, which that would have been 95 00:06:03,298 --> 00:06:14,140 RIFF of course, and that did very badly. You know, it very much 96 00:06:14,140 --> 00:06:21,646 underperformed the various other kinds of benchmark CTA funds 97 00:05:53,290 --> 00:05:58,450 So,you know, there are various explanations for this. Do 98 00:05:58,450 --> 00:06:04,642 they keep all the good stuff for themselves? Are they just good 99 00:06:04,642 --> 00:06:10,834 at one thing that has limited capacity and then the outside funds 100 00:06:10,834 --> 00:06:17,542 have to do other things? I don't know. But, yeah, I mean, 14% 101 00:06:17,542 --> 00:06:21,670 down in a month, It's a big number. 102 00:06:21,670 --> 00:07:12,347 I think, just to balance our conversation, I seem to remember 103 00:07:12,347 --> 00:08:03,024 that actually these funds have probably done pretty well the last 104 00:08:03,024 --> 00:09:07,522 few years. So, this is like, the journal is picking up on an outlier 105 00:09:07,522 --> 00:09:25,950 number. Let's be fair. ButI think you're right about the fact 106 00:06:50,014 --> 00:07:39,616 that they, at some point, also try to get into our industry. I haven't 107 00:07:39,616 --> 00:08:25,675 heard much about that ever since. But you know, we can all have 108 00:08:25,675 --> 00:09:08,191 an outlier month or an outlier year, without a doubt. So, we'll 109 00:09:08,191 --> 00:09:25,906 see whether it comes back. Theother thing that caught my attention, 110 00:06:42,474 --> 00:07:21,110 this is something I saw this morning. It was an article, and I 111 00:07:21,110 --> 00:08:02,718 don't even know how I found it. It was probably in one of those 112 00:08:02,718 --> 00:08:35,410 platforms that shows you all sorts of articles regarding or relating 113 00:08:35,410 --> 00:09:11,074 to our industry. It was an article on a webpage called CTOL. 114 00:09:11,074 --> 00:09:25,934 I never heard about it. Theheadline was Volatile Commodity 115 00:06:51,390 --> 00:07:32,998 Markets Test Quantitative Trading Funds Performance Claims. 116 00:07:32,998 --> 00:08:50,270 And that got my attention. One of the first sub headlines was something 117 00:08:50,270 --> 00:09:25,934 like, Performance Falls Short of Marketing. Andthey write about 118 00:06:45,446 --> 00:07:27,054 how presentations to institutional investors, that highlight 119 00:07:27,054 --> 00:08:20,550 the outperformance during turbulent times, like in 2022, is 120 00:08:20,550 --> 00:09:19,990 there. And it positions managed futures as an essential portfolio 121 00:09:19,990 --> 00:09:25,934 diversifier. Butit also talks about a Midwestern public pension 122 00:06:54,835 --> 00:07:31,685 fund being frustrated. It's quoting a female, probably CIO, being 123 00:07:31,685 --> 00:08:01,165 frustrated with CTA allocation. And they obviously added 124 00:08:01,165 --> 00:08:41,700 it in 2023 after the exceptional year in 2022. So, probably 125 00:08:41,700 --> 00:09:18,550 not the best timing and probably not enough time horizon 126 00:09:18,550 --> 00:09:25,920 as well. Butthen it went on to something that kind of really 127 00:06:42,559 --> 00:07:07,246 caught my attention. And this kind of warms up to next week's conversation, 128 00:07:07,246 --> 00:07:28,135 which will be with Andrew, because it also mentions the ETF 129 00:07:28,135 --> 00:07:49,024 offerings that, of course, have done very well this year. And 130 00:07:49,024 --> 00:08:11,812 it talks about, and I'm just quoting here, a newer category of 131 00:08:11,812 --> 00:08:27,004 managed futures exchange traded funds has performed better 132 00:08:27,004 --> 00:08:44,095 than many established CTA funds. These ETF packages, similar 133 00:08:44,095 --> 00:08:59,287 strategies, in cheaper, more transparent structures, are accessible 134 00:08:59,287 --> 00:09:22,075 to retail investors and they have seen a lot of inflows, blah 135 00:09:22,075 --> 00:09:25,873 blah blah. Andthen it goes on to say the irony is notable, products 136 00:06:50,386 --> 00:07:09,530 marketed less aggressively and with fewer institutional trappings 137 00:07:09,530 --> 00:07:31,067 have delivered much better results than prestigious legacy funds, 138 00:07:31,067 --> 00:08:04,569 blah blah blah. And I'm just going to say I don't think it's a 139 00:08:04,569 --> 00:08:33,285 fair comment to say that the ETF space is not being aggressive 140 00:08:33,285 --> 00:09:06,787 in their marketing. After all, we see them a lot, on all sorts of 141 00:09:06,787 --> 00:09:25,931 publications, including, of course, when Andrew is here. So,I 142 00:06:43,350 --> 00:08:42,590 thought that was kind of funny if that's the impression they've 143 00:08:42,590 --> 00:09:25,950 left with this organization. 144 00:09:25,950 --> 00:09:43,132 Yeah, I mean I'm always getting content asking me to invest 145 00:09:43,132 --> 00:09:58,752 in CTAs, but I suspect that's the various algorithms identifying 146 00:09:58,752 --> 00:10:14,372 me as somebody who would potentially be interested in that. 147 00:09:25,950 --> 00:09:33,210 Soyeah, just to tie up on the other point, I believe it was the 148 00:09:33,210 --> 00:09:38,050 Renaissance International Equity Fund, which I think is a sort 149 00:09:38,050 --> 00:09:42,890 of equity market neutral type fund, probably not dissimilar from 150 00:09:42,890 --> 00:09:49,182 the sort of thing that the likes of AQR offer, to be honest. 151 00:09:49,182 --> 00:09:55,958 And its worst year so far, I believe, in the past, in 2022, they 152 00:09:55,958 --> 00:10:03,218 were down 19%. So, you know, to be down, what was it, 7% for the 153 00:10:03,218 --> 00:10:09,510 year so far? It's not like sort of an outlier, but their performance, 154 00:10:09,510 --> 00:10:14,350 since inception, is only 3.7% so it's not really been... 155 00:10:14,380 --> 00:10:31,441 Yeah, I think, in fairness, again, just to balance the conversation 156 00:10:31,441 --> 00:10:51,604 here, I think I read something in the full article that at least 157 00:10:51,604 --> 00:11:11,767 one of these funds actually had a mandate to be not very correlated 158 00:11:11,767 --> 00:11:14,869 with equity. Soagain, we can't translate that just because, 159 00:10:31,660 --> 00:11:01,900 so far this year, the S&P 500 is doing well, that this fund has 160 00:11:01,900 --> 00:11:14,860 to do well. We don't know. Anyways,just a little bit of good 161 00:10:18,531 --> 00:10:27,426 news before we go on. And that is, of course, there is no AI bubble 162 00:10:27,426 --> 00:10:34,542 after all. And that is purely based on the fact that yesterday 163 00:10:34,542 --> 00:10:41,658 the world can stop worrying about an AI bubble, at least for 164 00:10:41,658 --> 00:10:47,588 now, because AI chip making linchpin Nvidia dropped a stellar 165 00:10:47,588 --> 00:10:52,925 earnings report yesterday that has, at least temporarily, eased 166 00:10:52,925 --> 00:11:00,041 concerns that the economy is on the verge of collapsing like the 167 00:11:00,041 --> 00:11:06,564 New York Mets in September, according to this article. I probably 168 00:11:06,564 --> 00:11:13,680 got it from Bloomberg, those nice words. Anyways, I have no idea 169 00:11:13,680 --> 00:11:14,866 about AI. 170 00:11:14,870 --> 00:11:17,550 I have no idea about the New York Mets either. 171 00:11:17,630 --> 00:11:19,190 Right, yeah, exactly. 172 00:11:19,190 --> 00:11:22,030 But I'm sure it'll be fine. 173 00:11:22,590 --> 00:11:32,160 The trend may continue a little bit longer. Anyways, speaking 174 00:11:32,160 --> 00:11:42,687 of trends, my trend barometer finished at 48 last night. That's 175 00:11:42,687 --> 00:11:57,042 neutral, I would say, but it is a bit weaker than it was 10 days 176 00:11:57,042 --> 00:12:09,483 ago. And I think performance, when we get to that in a second, 177 00:12:09,483 --> 00:12:20,010 reflects that. So far, in October, I was kind of optimistic 178 00:12:20,010 --> 00:12:31,494 after last week. I thought, pretty good start after all the first 179 00:12:31,494 --> 00:12:42,978 couple of days, not great in November, and then last week was 180 00:12:42,978 --> 00:12:52,548 actually decent. But this week, not so much. It's definitely 181 00:12:52,548 --> 00:13:04,989 a bit challenging, a bit of a headwind for managers, at least that's 182 00:13:04,989 --> 00:13:10,731 what I see on my side. Thereare a couple of things that 183 00:11:36,618 --> 00:12:02,670 are standing out, I guess, so far this month. It's still, you know, 184 00:12:02,670 --> 00:12:28,722 okay to be long in precious metals. It's okay in terms of the 185 00:12:28,722 --> 00:12:48,762 softs, they're doing okay, maybe even some of the products, 186 00:12:48,762 --> 00:13:08,802 oil products. But equities is probably where the real challenge 187 00:13:08,802 --> 00:13:10,806 is. So,depending on… And this is where also, you know, funds like 188 00:11:34,722 --> 00:11:46,854 ETFs and replicators and all of that stuff, it's all about exposure 189 00:11:46,854 --> 00:11:59,997 this year. It's all about how much equity have you got, or how 190 00:11:59,997 --> 00:12:13,140 much fixed income have you got, and so on, and so forth. So, 191 00:12:13,140 --> 00:12:24,261 to say conclusively that one approach is better than the other, 192 00:12:24,261 --> 00:12:35,382 I think is very dangerous this year. Because it's really depending 193 00:12:35,382 --> 00:12:46,503 on more, I think, your market universe than anything else right 194 00:12:46,503 --> 00:12:56,613 now. Maybe speed, of course, short-term has been challenged for 195 00:12:56,613 --> 00:13:07,734 sure. But market universe, this year, is really having a huge 196 00:13:07,734 --> 00:13:10,767 impact on returns. Someof the smaller currencies, New Zealand dollar, 197 00:11:47,326 --> 00:12:24,430 Mexican pesos, still doing okay as far as I can tell. Other 198 00:12:24,430 --> 00:12:58,442 than that, that's kind of been our experience. Anything from you 199 00:12:58,442 --> 00:13:10,810 in terms of takeaways? 200 00:13:13,070 --> 00:13:27,994 Yeah, yeah, sort of a similar picture for me. So, I'm pretty much 201 00:13:27,994 --> 00:13:41,770 flat for November, I guess. Yeah, like most people I had few 202 00:13:41,770 --> 00:13:56,694 good days and have given some of that back recently. For the year 203 00:13:56,694 --> 00:14:12,766 I'm up about sort of 5%. So, it's okay. It's not too bad. So, 204 00:14:12,766 --> 00:14:26,542 down about 3% from sort of mid October when things were looking 205 00:14:26,542 --> 00:14:27,690 good. AndI'm still in a sort of drawdown of about, well, just 206 00:13:29,966 --> 00:13:48,270 over 10%, I'd say. In terms of risk, kind of looking forward, so, 207 00:13:48,270 --> 00:14:02,350 interestingly, my biggest short is in bitcoin, which is very 208 00:14:02,350 --> 00:14:17,838 satisfying because you know my thoughts on that subject. But obviously 209 00:14:17,838 --> 00:14:27,694 that's the system hating it, not me. I'msort of net long equities, 210 00:13:18,212 --> 00:13:26,782 although for example, I'm short DAX, but I'm actually long 211 00:13:26,782 --> 00:13:35,352 the Italian equities, so there's a fair bit of dispersion 212 00:13:35,352 --> 00:13:46,493 there. But yeah, I guess my risk sizing is probably in line with 213 00:13:46,493 --> 00:13:55,920 your trend parameter. It's sort of definitely a bit below average. 214 00:13:55,920 --> 00:14:07,061 It's not the sort of… We’re going to get some fairly decent trends, 215 00:14:07,061 --> 00:14:16,488 and performance was really picking up kind of coming into sort 216 00:14:16,488 --> 00:14:25,915 of September, October, but things have gotten a little bit volatile 217 00:14:25,915 --> 00:14:27,629 since then. 218 00:14:28,260 --> 00:14:33,252 You mentioned bitcoin, so I can't help asking you, have you ever 219 00:14:33,252 --> 00:14:39,076 tried to model this cycle, it has a halving cycle, to see if there's 220 00:14:39,076 --> 00:14:40,740 any truth to it? 221 00:14:40,740 --> 00:14:43,785 Well, as a quant, I mean, I'm not sure, off the top of my head, 222 00:14:43,785 --> 00:14:45,003 how many halvings there have been. 223 00:14:45,020 --> 00:14:46,420 There's been like four. 224 00:14:46,900 --> 00:14:49,793 Okay, so that's not really enough data points for a statistically 225 00:14:49,793 --> 00:14:52,949 significant model, is it? And you should know better than to even 226 00:14:52,949 --> 00:14:54,264 ask me that question, Niels. 227 00:14:54,590 --> 00:14:57,060 Well, anything to do with crypto, I have to kind of just poke 228 00:14:57,060 --> 00:14:57,630 a little bit. 229 00:14:58,190 --> 00:15:00,430 You're trying to make me hot. I know. You're trying to get me excited. 230 00:15:00,430 --> 00:15:00,910 Exactly. 231 00:15:00,910 --> 00:15:01,510 Trying to make me hot. 232 00:15:01,510 --> 00:15:27,730 Yeah, good, all right, good. All right, as of Tuesday, not so 233 00:15:27,730 --> 00:15:51,765 hot, I have to say, the performance numbers. Down 62 basis 234 00:15:51,765 --> 00:16:20,170 points for the BTOP 50, down 60 basis points for the year. SocGen 235 00:16:20,170 --> 00:16:48,575 CTA index down 69 basis points for the month, down 2.2% for the 236 00:16:48,575 --> 00:17:14,795 year. The SocGen Trend index is down 30 basis points, down 1.18% 237 00:17:14,795 --> 00:17:36,645 for the year. And the Short-Term Traders index down 88 238 00:17:36,645 --> 00:18:00,680 basis points, down 5.16% for the year. Yesterday probably a mixed 239 00:18:00,680 --> 00:18:33,455 day, some up, some down, so it could be a little bit worse as of 240 00:18:33,455 --> 00:18:35,640 today. MSCIWorld also down, down 2.79% so far this month and 241 00:15:42,826 --> 00:16:24,142 that is as of yesterday, Wednesday, and it's up 16.86% for 242 00:16:24,142 --> 00:17:12,970 the year. The S&P US Aggregate Bond Index down 10 basis points in 243 00:17:12,970 --> 00:18:05,554 November, up 6.5 for the year. And the S&P 500 down 2.82% as of 244 00:18:05,554 --> 00:18:35,602 last night, up 14.2% so far this year. Now,we've got this wonderful 245 00:15:17,980 --> 00:16:04,096 topic that we're going to keep a little bit longer in terms of suspense, 246 00:16:04,096 --> 00:16:40,330 but we've got lots of great questions so we appreciate that. 247 00:16:40,330 --> 00:17:19,858 Thank you for sending them in. Now, I've not filtered them, I'm 248 00:17:19,858 --> 00:18:05,974 just going to take them as they came in in the order they came 249 00:18:05,974 --> 00:18:35,620 in and read them as best as I can. Allright, first one is from 250 00:15:13,294 --> 00:15:32,934 George. My question is what does a solid reliable robustness 251 00:15:32,934 --> 00:15:52,574 testing framework look like? I'm struggling in the ‘try different 252 00:15:52,574 --> 00:16:12,214 things out’ stage e.g. too robust, every strategy fails even 253 00:16:12,214 --> 00:16:35,782 though they look good on unseen OOS data. And if too lose 254 00:16:35,782 --> 00:16:57,386 we yield a fair few good looking strategies e.g. my current 255 00:16:57,386 --> 00:17:18,990 idea: build strategy on three to five years of data, depending 256 00:17:18,990 --> 00:17:42,558 on time frame, and therefore risk of regime change. Do a quick 257 00:17:42,558 --> 00:18:04,162 high precision test on the build data. Those that do significant 258 00:18:04,162 --> 00:18:25,766 worse bin them, then a slippage test. Do they still make 259 00:18:25,766 --> 00:18:35,640 money if slippage is increased? Ifindthis particularly 260 00:15:16,802 --> 00:16:02,678 harsh assuming you get slipped five pips on every trade. Monte Carlo 261 00:16:02,678 --> 00:16:29,439 testing, randomize trade order, permutate strategy parameters, 262 00:16:29,439 --> 00:17:11,492 not optimize but see if changing parameters by 10% up or 263 00:17:11,492 --> 00:17:53,545 down breaks the strategy. I believe this helps to avoid curve 264 00:17:53,545 --> 00:18:35,598 fitting in the mining process. Then we test on tick data. Bythis 265 00:15:10,620 --> 00:16:00,725 point most strategies still in the running seem to survive this. 266 00:16:00,725 --> 00:16:46,275 Performance looks good and ranking of the best strategies changed 267 00:16:46,275 --> 00:17:40,935 slightly. Then the ultimate test is on unseen OOS tick data 1 268 00:17:40,935 --> 00:18:35,595 to 3 years, depending on the data sample in the build stage. Thenwe 269 00:15:06,862 --> 00:15:33,622 have, hopefully, a good few strategies to incubate which hope 270 00:15:33,622 --> 00:16:00,382 leads to profitable live trading. Please can you guys comment 271 00:16:00,382 --> 00:16:32,494 discuss the process. I find my robustness test may be too strict. 272 00:16:32,494 --> 00:17:04,606 All my strategies are just BEEP (I'm just going to say that 273 00:17:04,606 --> 00:17:42,070 wordless BEEP.) But I always have in mind, how do I know I filtered 274 00:17:42,070 --> 00:18:11,506 out the good strategies that will last. My process is strictly 275 00:18:11,506 --> 00:18:35,590 and yields a few excellent strategies. Thanks in advance. Okay,that 276 00:16:02,690 --> 00:18:35,640 was a very long thing. 277 00:18:35,640 --> 00:23:44,884 Just to be clear, BEEP is a bad word. He doesn't think these 278 00:23:44,884 --> 00:25:43,824 are good strategies, so yeah. There'sactually a big piece of information 279 00:19:01,036 --> 00:19:37,316 missing from this, which is what your expected holding period 280 00:19:37,316 --> 00:20:20,852 is because that's going to make quite a big difference to whether 281 00:20:20,852 --> 00:21:04,388 this process makes sense. So, I would say, for example, if you 282 00:21:04,388 --> 00:21:44,296 were trading with very short time frames, so intraday, then the 283 00:21:44,296 --> 00:22:27,832 timeframes and the things in the process kind of make sense. Although, 284 00:22:27,832 --> 00:23:14,996 I would say, if you are doing that, then you should probably start 285 00:23:14,996 --> 00:23:51,276 with higher frequency data because he talks about build data 286 00:23:51,276 --> 00:24:34,812 and then tick data. So, if you're trading on a fast timeframe, 287 00:24:34,812 --> 00:25:14,720 you really ought to be starting with tick data; with data 288 00:25:14,720 --> 00:25:43,744 that's much faster than your holding period is. Onthe other hand, 289 00:19:03,716 --> 00:20:27,944 if your holding period is more analogous to mine, which is about 290 00:20:27,944 --> 00:22:06,210 say an average of a month, then you don't need to go near tick 291 00:22:06,210 --> 00:23:37,457 data. There's absolutely no need to go near it at all. You can 292 00:23:37,457 --> 00:25:08,704 just do simple things to sort of check what the effect of your 293 00:25:08,704 --> 00:25:43,799 execution is on your performance. Andif you're holding 294 00:18:50,032 --> 00:19:33,208 periods a month, then even delaying your trades by a day (which 295 00:19:33,208 --> 00:20:09,188 is what I test in my backtester) shouldn't change your 296 00:20:09,188 --> 00:20:52,364 results that much. And if it does, you've got a problem somewhere. 297 00:20:52,364 --> 00:21:42,736 But on the other hand, if you are holding for ,say, a month, then 298 00:21:42,736 --> 00:22:33,108 I would say that three to five years probably, for me, is an insufficient 299 00:22:33,108 --> 00:23:16,284 amount of time to do sort of the calibration testing of strategies. 300 00:23:16,284 --> 00:24:10,254 And I know he then has another one to three years of out of sample. 301 00:24:10,254 --> 00:25:00,626 So, I think it feels to me like you might not be getting sort 302 00:25:00,626 --> 00:25:43,802 of robust results because of this sort of brief periods of time. 303 00:18:35,640 --> 00:20:22,684 Justto say in advance, by the way, overall there's no kind of massive 304 00:20:22,684 --> 00:21:54,436 red flags in here; all the steps individually kind of make sense. 305 00:21:54,436 --> 00:23:18,542 I think it's more whether together they make sense, and again, 306 00:23:18,542 --> 00:24:50,294 whether the sort of time periods being used and the data being 307 00:24:50,294 --> 00:25:43,816 used matches the strategy that you're trading. Anyway,so, putting 308 00:18:53,991 --> 00:20:01,278 that aside then, I think this is interesting stuff actually because 309 00:20:01,278 --> 00:21:14,682 (not the book I'm currently writing, but the book I'm going to 310 00:21:14,682 --> 00:22:28,086 write after this one) I'm going to write a book on backtesting 311 00:22:28,086 --> 00:23:29,256 because I think there's partly a misunderstanding, but partly there's 312 00:23:29,256 --> 00:24:48,777 a sort of deep conflict between what the point is of a backtest 313 00:24:48,777 --> 00:25:43,830 and how you should use the results of it. Becausebacktests 314 00:18:45,714 --> 00:19:36,084 fulfill two functions. They fulfill the function of finding you 315 00:19:36,084 --> 00:20:36,528 a strategy that's going to work well in the future, and they 316 00:20:36,528 --> 00:21:36,972 also tell you how well you would have performed in the past. 317 00:21:36,972 --> 00:22:27,342 And ideally, they would do both. And this particular framework 318 00:22:27,342 --> 00:23:42,897 is not going to do that for you. It's not going to tell you how 319 00:23:42,897 --> 00:24:43,341 well he would have done in the past. Because you're basically selecting 320 00:24:43,341 --> 00:25:43,785 the strategies that did the best with the data that you've got. 321 00:18:35,640 --> 00:19:26,236 Andyou've got this in sample and out of sample process. But that's 322 00:19:26,236 --> 00:20:16,832 only going part of the way to doing that. Because the real kind 323 00:20:16,832 --> 00:21:11,320 of gold standard way of doing this is you start with, say, a thousand 324 00:21:11,320 --> 00:21:54,132 strategies and you basically do a rolling process where you sub 325 00:21:54,132 --> 00:22:29,160 select from those strategies depending on the performance you've 326 00:22:29,160 --> 00:23:19,756 only seen in the past. And you'll be doing that over the eight 327 00:23:19,756 --> 00:24:06,460 year period. And at the start of that eight year period you've 328 00:24:06,460 --> 00:24:49,272 got no information. So, you just include all the strategies you 329 00:24:49,272 --> 00:25:39,868 back test, including ones that are BEEP as well as ones that are 330 00:25:39,868 --> 00:25:43,760 good. Andyour performance initially would be pretty, I mean, 331 00:19:23,214 --> 00:20:31,932 let's say on average half your strategy is half is BEEP, then your 332 00:20:31,932 --> 00:21:30,078 performance, to begin with, will be flat because you'd have good 333 00:21:30,078 --> 00:22:12,366 performance and bad performance and they cancel each 334 00:22:12,366 --> 00:23:10,512 other out. Over time you would hopefully discover what the good 335 00:23:10,512 --> 00:24:13,944 strategies are and you would include more of those in your portfolio 336 00:24:13,944 --> 00:25:17,376 blend, and you'd end up with the set of strategies that you'd 337 00:25:17,376 --> 00:25:43,806 want to trade going forwards. Andthen, if you look at that historic 338 00:19:00,456 --> 00:19:31,476 backtest to performance, that's telling you how well you would 339 00:19:31,476 --> 00:20:08,700 actually have done had you had no future information at the start 340 00:20:08,700 --> 00:20:49,026 of your backtest. And it's also going to give you, at the end 341 00:20:49,026 --> 00:21:29,352 of the back test, it's going to give you the mix of strategies 342 00:21:29,352 --> 00:21:57,270 that you should trade going forwards. But importantly, doing 343 00:21:57,270 --> 00:22:34,494 it this way means that it's much more likely that, if you're 344 00:22:34,494 --> 00:23:05,514 doing your sort of strategy selection in a statistically robust 345 00:23:05,514 --> 00:23:36,534 manner, you're much less likely to do something like (and 346 00:23:36,534 --> 00:24:16,860 it sounds like this guy's doing it) pick, say, 1% out of those 347 00:24:16,860 --> 00:24:54,084 thousand strategies; to pick just 10 that do the very best. Because 348 00:24:54,084 --> 00:25:25,104 it's extremely unlikely unless, as I said, you are trading 349 00:25:25,104 --> 00:25:43,716 at a very short time frame. It'sextremely unlikely that after 350 00:18:54,420 --> 00:19:50,760 3, 4, 5, 6, 7, 8 years of time that enough time will have passed 351 00:19:50,760 --> 00:20:32,076 you to be able to say, with statistical confidence, that these 352 00:20:32,076 --> 00:21:13,392 10 strategies are the best and you shouldn't trade anything else. 353 00:21:13,392 --> 00:21:58,464 It’s much more likely that you'd end up with… So, you know, 354 00:21:58,464 --> 00:22:36,024 you're kind of fitting weights to a thousand different strategies. 355 00:22:36,024 --> 00:23:24,852 Maybe half of them you'll kind of have enough evidence to get rid 356 00:23:24,852 --> 00:24:13,680 of entirely, but most of the rest would still be in there. Yes, 357 00:24:13,680 --> 00:24:58,752 your 10 superstars would have a higher weight than the other others, 358 00:24:58,752 --> 00:25:43,824 but, by no means, would they be filling the entire portfolio up. 359 00:18:35,640 --> 00:19:10,785 Andthe issue here is this thing called the multiple testing 360 00:19:10,785 --> 00:19:49,125 problem. And essentially, even if you set a very high bar, let's 361 00:19:49,125 --> 00:20:37,050 say you say, well, I only want to see the top 1%, if you've got 362 00:20:37,050 --> 00:21:15,390 a thousand things, then just by luck, even if on average they're 363 00:21:15,390 --> 00:21:56,925 kind of flat performance, just by luck, ten of them are going to 364 00:21:56,925 --> 00:22:38,460 be amazing. And then you pick those 10 out. And yes, he's doing 365 00:22:38,460 --> 00:23:10,410 these other robustness tests that are good and will hopefully 366 00:23:10,410 --> 00:23:51,945 kind of weed out anything too crazy. But you know, I still think 367 00:23:51,945 --> 00:24:27,090 that potentially the main issue with this is that you're selecting 368 00:24:27,090 --> 00:24:55,845 a very small subset of strategies based on insufficient 369 00:24:55,845 --> 00:25:37,380 data. That's kind of the big red flag for me with this process 370 00:25:37,380 --> 00:25:43,770 as described. But,you know, doing things like Monte Carlo testing, 371 00:19:27,714 --> 00:20:37,146 doing things like holding out a sample, that's all good, but you 372 00:20:37,146 --> 00:21:52,364 know… Sorry, the one other thing to say is as well, you know, 373 00:21:52,364 --> 00:23:13,368 he tests stuff and then tests to see what the effect is of adding 374 00:23:13,368 --> 00:24:17,014 trading costs. Trading costs should be in there from the very 375 00:24:17,014 --> 00:25:14,874 start. There's no point even considering things that you can't 376 00:25:14,874 --> 00:25:43,804 make money with before slippage. So,you know, unless there's 377 00:19:38,610 --> 00:21:44,550 something about your back testing process that makes it very 378 00:21:44,550 --> 00:23:37,896 computationally expensive to include costs, and that's why you're 379 00:23:37,896 --> 00:25:43,836 doing this, I don't understand otherwise why you do that. So,that's 380 00:18:48,234 --> 00:20:16,392 kind of my general thoughts. And, as I said, I could go on and 381 00:20:16,392 --> 00:21:44,550 you know, as I said, if I'm going to write a book about this, 382 00:21:44,550 --> 00:23:00,114 which will probably be at least 300 pages, then I could clearly 383 00:23:00,114 --> 00:24:21,975 go on for a long time to discuss this topic. But that's sort 384 00:24:21,975 --> 00:25:31,242 of my thinking about how you should backtest and the purposes 385 00:25:31,242 --> 00:25:43,836 of backtest. Andhopefully that gives you an idea of some of 386 00:21:59,540 --> 00:25:43,830 the changes you can make to this process to Improve it. 387 00:25:46,261 --> 00:25:52,111 It's all well and good to follow a process and you know, you 388 00:25:52,111 --> 00:25:57,061 end up deciding on something you feel comfortable with. But things 389 00:25:57,061 --> 00:26:02,461 change, data changes. How often would you go back and do this, 390 00:26:02,461 --> 00:26:06,961 and would you simply change your, say, models, parameters, whatever, 391 00:26:06,961 --> 00:26:12,361 on an ongoing basis in the future, based on future tests that 392 00:26:12,361 --> 00:26:13,261 you make? 393 00:26:14,731 --> 00:26:23,981 Well, basically, in theory, you run the process forward. So, 394 00:26:23,981 --> 00:26:35,081 if your back testing process consists of… So, if you've got a 395 00:26:35,081 --> 00:26:44,331 monthly holding period, then I think it's probably appropriate to 396 00:26:44,331 --> 00:26:52,656 do this re-optimization of weights to strategies about every 397 00:26:52,656 --> 00:27:03,756 year, say. There's no point in doing it more frequently than that. 398 00:27:03,756 --> 00:27:13,931 And that will then result in your weight slowly evolving. And 399 00:27:13,931 --> 00:27:26,881 depending on your time frame, you know, you will get to see a slow 400 00:27:26,881 --> 00:27:36,131 evolution. Some things will become better, some things will become 401 00:27:36,131 --> 00:27:45,381 worse. So, that, in theory, should just roll that forward. So,at 402 00:26:17,117 --> 00:26:32,626 AHL, we used to have an annual refit of our trading strategies where 403 00:26:32,626 --> 00:26:45,749 basically we do exactly that, we'd include another year of data. 404 00:26:45,749 --> 00:27:01,258 But if you think about it, if you've got a long enough history, 405 00:27:01,258 --> 00:27:19,153 so you know, my data goes back to 1970, so that's now 55 years of 406 00:27:19,153 --> 00:27:34,662 data. At the beginning of next year, I'll have an extra year. I'm 407 00:27:34,662 --> 00:27:45,399 adding less than 2% to my total data set. Nowyou could argue 408 00:26:18,151 --> 00:26:27,556 that you should weight more recent data more highly than that, 409 00:26:27,556 --> 00:26:37,816 but even so, if you've got long back tests and relatively slow 410 00:26:37,816 --> 00:26:45,511 trading, then actually your weights probably aren't going to 411 00:26:45,511 --> 00:26:54,916 change that much. And the truth is, for me personally, I've 412 00:26:54,916 --> 00:27:04,321 refitted my weights on average about every three years. So, you 413 00:27:04,321 --> 00:27:15,436 know, every year seems like a lot of work. So, normally only if 414 00:27:15,436 --> 00:27:26,551 I make other changes to my system I then go and also change 415 00:27:26,551 --> 00:27:35,956 the weights. Because, as I said, with such long holding periods, 416 00:27:35,956 --> 00:27:45,361 there's not really evidence to say you should do anything differently. 417 00:27:45,681 --> 00:28:08,853 No, I think that's a perfect follow up answer. Thanks for that. 418 00:27:45,681 --> 00:27:50,313 Allright, we move on to a question from Frederick. He writes, 419 00:27:50,313 --> 00:27:55,331 hello, thanks for the show. I've learned a lot from it. My question 420 00:27:55,331 --> 00:27:59,963 for Rob is if he has any thoughts or statistics around intraday 421 00:27:59,963 --> 00:28:03,823 versus end-of-day trading for a medium to long-term trend system. 422 00:28:03,823 --> 00:28:07,683 Would the faster entries and tighter risk control actually lead 423 00:28:07,683 --> 00:28:08,841 to better results? 424 00:28:11,421 --> 00:29:25,443 I am a bit confused by this question to be honest, because the 425 00:29:25,443 --> 00:30:33,771 two sentences to me don't seem related to each other. So, let's 426 00:30:33,771 --> 00:31:36,405 say you're trading medium to long-term trends, okay? So, you can 427 00:31:36,405 --> 00:32:10,569 trade that system in different ways. So,one way you could trade 428 00:28:32,835 --> 00:29:15,663 it is, basically, a lot of people backtest their stuff on end-of-day 429 00:29:15,663 --> 00:29:51,353 data, which is completely fine if you're trading slowly enough, 430 00:29:51,353 --> 00:30:34,181 as we discussed the first question. So, one thing you can do 431 00:30:34,181 --> 00:31:06,302 is, basically, then, when you're trading, actually try and 432 00:31:06,302 --> 00:31:56,268 trade as close to the end of the day as possible. So, you sort 433 00:31:56,268 --> 00:32:10,544 of matching your backtest. Nowthat might not be ideal. And 434 00:28:31,350 --> 00:29:02,667 it's funny actually because I've done consulting and a lot of 435 00:29:02,667 --> 00:29:39,678 people have got this sort of fixation with oh, I must trade at 436 00:29:39,678 --> 00:30:13,842 the end of the day because that's where my backtest data's from. 437 00:30:13,842 --> 00:30:50,853 And I said to them, well just try this experiment which I already 438 00:30:50,853 --> 00:31:22,170 mentioned, just delay your fills by a day in your backtest. 439 00:31:22,170 --> 00:31:53,487 Does it make it any difference? Does it not make any 440 00:31:53,487 --> 00:32:10,569 difference? Okay, well then you're fine. Oryou know, as long 441 00:28:22,001 --> 00:28:49,509 as you do something like what I do, which is essentially, you know, 442 00:28:49,509 --> 00:29:10,669 take the previous day's closing prices and then, at some 443 00:29:10,669 --> 00:29:31,829 point during the following day, do your trading, you'll know 444 00:29:31,829 --> 00:29:48,757 you're going to achieve performance somewhere between if 445 00:29:48,757 --> 00:30:16,265 you'd managed to do it exactly at the close of the previous day, 446 00:30:16,265 --> 00:30:39,541 which you can't, and the following day is closed, which is 447 00:30:39,541 --> 00:31:02,817 what I backtest, you're going to be somewhere between those two. 448 00:31:02,817 --> 00:31:26,093 And if those two things are close enough, then obviously, on 449 00:31:26,093 --> 00:31:49,369 a day-to-day basis, you might see some differences of prices have 450 00:31:49,369 --> 00:32:10,529 been a bit crazy but on average you'll be fine. So,I would 451 00:28:17,715 --> 00:28:38,695 suggest… And the other advantage of that, of course, is 452 00:28:38,695 --> 00:29:08,067 if the close is not liquid, or is crazy, or there's some weird auction 453 00:29:08,067 --> 00:29:31,145 process, or whatever, you can focus your trading at other parts 454 00:29:31,145 --> 00:29:56,321 of the day that are more liquid. And, of course, if you're 455 00:29:56,321 --> 00:30:17,301 trading institutional size money, the last thing you want to 456 00:30:17,301 --> 00:30:44,575 do is do one massive trade every day. You want to spread your 457 00:30:44,575 --> 00:31:07,653 trades throughout the day and try and reduce your market impact 458 00:31:07,653 --> 00:31:30,731 as much as possible. So, if you're trading medium to long-term 459 00:31:30,731 --> 00:31:53,809 trends where, basically, you want to trade roughly once a day, 460 00:31:53,809 --> 00:32:10,593 then spreading that out is no bad thing. Theother thing you can 461 00:28:21,816 --> 00:28:44,685 do, if you have a system like mine which essentially dynamically 462 00:28:44,685 --> 00:29:09,633 looks at its positions sort of optimal versus what it actually has, 463 00:29:09,633 --> 00:29:32,502 I could actually run that thing throughout the day. And that 464 00:29:32,502 --> 00:29:59,529 would not increase the total amount of trading I did, but it would, 465 00:29:59,529 --> 00:30:24,477 again, result in a sort of smoothing process and it would allow 466 00:30:24,477 --> 00:30:47,346 me to react faster to very large intraday price movements which 467 00:30:47,346 --> 00:31:14,373 may or may not be a good thing because, obviously, if prices bounce 468 00:31:14,373 --> 00:31:37,242 back, then that's actually going to lose you money. So that's 469 00:31:37,242 --> 00:32:04,269 kind of my answer to what is, for me, quite a confusing question, 470 00:32:04,269 --> 00:32:10,506 to be honest. AndI don't really understand what faster entries 471 00:28:48,933 --> 00:29:31,134 and tighter risk control actually mean unless he's talking 472 00:29:31,134 --> 00:30:13,335 about trading much more quickly. So, he's actually talking 473 00:30:13,335 --> 00:31:09,603 about sort of intraday trend trading. So, you know, in which case, 474 00:31:09,603 --> 00:32:05,871 you know, I've got two big issues with people who trade trends 475 00:32:05,871 --> 00:32:10,560 intraday. Oneis obviously costs. Costs are going to be higher 476 00:28:47,111 --> 00:29:22,801 because you're doing more trades, but also because it's harder 477 00:29:22,801 --> 00:30:05,629 to do your trades with sort of passive limit orders because you 478 00:30:05,629 --> 00:30:55,595 want to catch a trend that's only going to last a couple of hours. 479 00:30:55,595 --> 00:31:45,561 You don't want to sort of wait there for it to be passively executed. 480 00:31:45,561 --> 00:32:10,544 You need to jump in straight away. Andthe other issue is if you 481 00:28:28,858 --> 00:28:58,750 actually look at the efficacy of trend systems, they tend to start 482 00:28:58,750 --> 00:29:31,133 to decline for holding periods shorter than about a month or a couple 483 00:29:31,133 --> 00:30:01,025 of weeks, depending on the asset class. And by the time you 484 00:30:01,025 --> 00:30:28,426 get down to intraday, it's less obvious that there are strong 485 00:30:28,426 --> 00:30:55,827 trend patterns there. And, actually, if you get down to sort 486 00:30:55,827 --> 00:31:28,210 of sub one hour, you tend to start to see mean reversion, which 487 00:31:28,210 --> 00:31:58,102 obviously is the opposite of trend. So, I'd be very careful about 488 00:31:58,102 --> 00:32:10,557 looking at intraday trend trading. Fasterentries and tighter 489 00:28:51,281 --> 00:30:50,861 risk control, whatever that means, aren't going to be a lot of 490 00:30:50,861 --> 00:32:10,581 help when you're paying those big trading costs. 491 00:32:11,081 --> 00:32:44,179 I'm thinking that, you know, if you are newer to the space, and 492 00:32:44,179 --> 00:33:07,093 you get maybe information from, you know, online platforms, 493 00:33:07,093 --> 00:33:32,553 people promoting stuff, etc., etc., I can understand why people 494 00:33:32,553 --> 00:34:03,105 might think that actually, oh, trading faster, as long as it's trend, 495 00:34:03,105 --> 00:34:08,197 it works. AndI even had a conversation on email from a nice 496 00:32:21,146 --> 00:32:32,126 guy in the US regarding using trend following just on a single 497 00:32:32,126 --> 00:32:42,191 asset class. They're doing it on equities. And I'm thinking, okay, 498 00:32:42,191 --> 00:32:54,086 it's great if it works for them. But it's just not my experience 499 00:32:54,086 --> 00:33:01,406 that, actually, trend following isn’t designed to work 500 00:33:01,406 --> 00:33:12,386 either on faster speeds or on a single asset class. It's really 501 00:33:12,386 --> 00:33:21,536 the diversification, and the longer holding period, and all that 502 00:33:21,536 --> 00:33:32,516 stuff, that makes it a little bit difficult, makes it a little 503 00:33:32,516 --> 00:33:43,496 bit uncomfortable, that makes it work. But I do take that people 504 00:33:43,496 --> 00:33:52,646 might have a different impression coming to the space and 505 00:33:52,646 --> 00:34:01,796 maybe getting a lot of the information initially from, say, 506 00:34:01,796 --> 00:34:08,201 more commercial outlets, let's call it that. Anyways,we appreciate 507 00:33:09,641 --> 00:34:08,201 the question, Frederick. Allright, question from Abram. I'm 508 00:32:17,771 --> 00:32:28,921 starting to read Mr. Robert Carver's book on systematic trading 509 00:32:28,921 --> 00:32:41,186 and found his trading journal in elite trading, spanning from February 510 00:32:41,186 --> 00:32:54,566 of 2015 until now. After reading the first 10 pages, I skipped 511 00:32:54,566 --> 00:33:07,946 to the most recent pages and saw Mr. Carver inviting readers to 512 00:33:07,946 --> 00:33:20,211 ask questions in your upcoming interview. Can you please help to 513 00:33:20,211 --> 00:33:31,361 ask how Mr. Carver will start his automated systematic trading 514 00:33:31,361 --> 00:33:42,511 differently? Given over 10 years of experience, what kind of 515 00:33:42,511 --> 00:33:55,891 advice can Mr. Carver give to a new systematic trader just starting 516 00:33:55,891 --> 00:34:08,156 out in this area? Thank you and best regards from Abram. 517 00:34:08,841 --> 00:34:12,201 I mean, it's quite depressing that you can only manage 10 pages 518 00:34:12,201 --> 00:34:15,561 of… It's not clear whether it's my trading journal or my book, 519 00:34:15,561 --> 00:34:18,921 only 10 pages before he just grew so bored and just skipped. 520 00:34:18,951 --> 00:34:21,447 I think he did the right thing. He went straight to the source 521 00:34:21,447 --> 00:34:22,983 and asked his questions. I think he did. 522 00:34:22,991 --> 00:35:03,969 Yeah, well, if this is a ploy to try and avoid reading the rest 523 00:35:03,969 --> 00:35:39,093 of the book or the, or the trading journal, tough luck, you're 524 00:35:39,093 --> 00:36:17,144 still going to have to do that. So, that's my first piece of 525 00:36:17,144 --> 00:36:43,487 advice. Read till the end. It's all good stuff. Although,to 526 00:34:32,681 --> 00:35:30,821 be fair, the trading journal is many thousands of posts long now, 527 00:35:30,821 --> 00:36:33,806 so, you know, it'll take a while. The book's only a few hundred 528 00:36:33,806 --> 00:36:43,496 pages though. Yeah,it's an interesting question actually because, 529 00:35:24,458 --> 00:36:43,487 you know, what's changed over the last ten years? It'sactually… 530 00:34:26,183 --> 00:34:46,931 Where are we? It's about eleven and a half years since I started 531 00:34:46,931 --> 00:35:01,295 running my own automated trading platform. And it's nearly 532 00:35:01,295 --> 00:35:18,851 twenty years since I began working in the CTA industry. So, 533 00:35:18,851 --> 00:35:38,003 you know, quite a few things probably have changed in that time 534 00:35:38,003 --> 00:35:52,367 period. But I'm actually struggling. I mean, Niels, you've 535 00:35:52,367 --> 00:36:11,519 been around even longer than I have. Although people might not guess 536 00:36:11,519 --> 00:36:30,671 it because you still have a reasonable amount of hair. So, you 537 00:36:30,671 --> 00:36:43,501 probably look about ten years younger than me. Imean,I don't 538 00:34:26,735 --> 00:34:37,031 know what you think, but I'm actually struggling to think how 539 00:34:37,031 --> 00:34:48,263 my advice would have changed from ten or even twenty years ago. 540 00:34:48,263 --> 00:35:01,367 I think, I probably say the same things. I mean, if I was to 541 00:35:01,367 --> 00:35:13,535 go and revisit that book I wrote, over 10 years ago now, I'm 542 00:35:13,535 --> 00:35:25,703 not sure very much in it would change. I mean, I'd probably put 543 00:35:25,703 --> 00:35:35,999 some stuff in for crypto, just that specific asset class that's 544 00:35:35,999 --> 00:35:48,167 kind of come out of nowhere. And my new book that I'm writing 545 00:35:48,167 --> 00:36:00,335 now is, has got some stuff on crypto just for that reason. But 546 00:36:00,335 --> 00:36:11,567 apart from that, you know, I'm not really feeling that there's been… 547 00:36:11,567 --> 00:36:22,799 you know… I think the same principles apply. I mean things we 548 00:36:22,799 --> 00:36:32,159 talked about like backtesting. I think those are principles that 549 00:36:32,159 --> 00:36:43,391 have applied and it's not like they're sort of new scientific discoveries. 550 00:34:22,991 --> 00:34:47,783 Theadvice I normally give to people is things like keep things 551 00:34:47,783 --> 00:35:06,377 simple. As we've discussed, think about trading costs, think 552 00:35:06,377 --> 00:35:24,971 about leverage, optimal leverage. Don't take too much risk 553 00:35:24,971 --> 00:35:45,631 because the three things that kill systematic traders are spending 554 00:35:45,631 --> 00:36:08,357 too much on costs, taking too much leverage, and overfitting their 555 00:36:08,357 --> 00:36:33,149 trading systems. And those are the same three things I was warning 556 00:36:33,149 --> 00:36:43,479 about 10, 12 years ago. Andthose are the same things that 557 00:34:48,870 --> 00:35:36,931 I think were a problem 20 years ago. So yeah, I'm going to 558 00:35:36,931 --> 00:36:21,295 hand this one back to you Niels, as you’re even more experienced 559 00:36:21,295 --> 00:36:43,477 than me, what do you think? 560 00:36:43,741 --> 00:36:48,496 So, first of all I would say… And I think this was your first book, 561 00:36:48,496 --> 00:36:51,032 if I'm not mistaken, Systematic Trading Strategies. Am 562 00:36:51,032 --> 00:36:54,836 I correct in saying that? I think that was, actually, the first 563 00:36:54,836 --> 00:36:58,640 time you were on the podcast was when we reviewed that book. 564 00:36:58,661 --> 00:36:59,781 It was, yeah. 565 00:36:59,781 --> 00:37:38,523 Which is a great book and everybody should really read it, 566 00:37:38,523 --> 00:38:31,353 and it is timeless, and I tend to agree with you that not a lot 567 00:38:31,353 --> 00:39:13,617 of change. Where I think things may have changed is that I 568 00:39:13,617 --> 00:39:52,359 don't necessarily feel that the way we try to identify entries 569 00:39:52,359 --> 00:40:34,623 would have changed. Of course we still adhere to the various golden 570 00:40:34,623 --> 00:41:06,321 rules of trend following that have lasted for decades. Idothink, 571 00:37:12,987 --> 00:38:05,811 maybe, the industry has evolved when it comes to how we manage 572 00:38:05,811 --> 00:38:58,635 risk. It's difficult for me to say specifically how each firm would 573 00:38:58,635 --> 00:40:09,067 have evolved, but if I look at what we do at Dunn, I would say we 574 00:40:09,067 --> 00:41:06,293 do things differently. We do it in a smarter way, in a sense. 575 00:36:59,781 --> 00:41:06,321 Ithink,also, we are more open to differentiating speeds a 576 00:37:21,297 --> 00:37:40,857 bit, and I don't mean just deploying short term strategies. 577 00:37:40,857 --> 00:38:06,285 That's not what I mean. But you can design your systems to react 578 00:38:06,285 --> 00:38:27,801 differently from entries to exits. I think that that is something 579 00:38:27,801 --> 00:38:55,185 we do, and I think that has helped us. And as I said, overall 580 00:38:55,185 --> 00:39:16,701 risk management, I feel, has evolved as well. Maybe people like 581 00:39:16,701 --> 00:39:40,173 you wrote another blog post, which we're not going to talk about 582 00:39:40,173 --> 00:40:03,645 today, but that Katy plugged for you last week, was about predicting 583 00:40:03,645 --> 00:40:21,249 volatility. Of course, volatility…. Maybe we are using volatility 584 00:40:21,249 --> 00:40:46,677 differently today than we did 20 years ago and maybe we can use 585 00:40:46,677 --> 00:41:06,237 it better now in the way we manage the risk. So,I generally 586 00:37:08,799 --> 00:37:47,877 agree with you that it's not massive changes. But I do think that 587 00:37:47,877 --> 00:38:23,949 we have evolved. I mean, there's a reason why we have these 588 00:38:23,949 --> 00:39:00,021 research teams who are very good at what they do. It doesn't 589 00:39:00,021 --> 00:39:30,081 mean that they can overcome bad market environments. And we've 590 00:39:30,081 --> 00:40:12,165 had some of that in the last 20 years for sure. And that's why 591 00:40:12,165 --> 00:40:42,225 people sometimes write that, oh, this strategy doesn't work as 592 00:40:42,225 --> 00:41:06,273 well anymore. I'm not so sure that's true. Atthe end of the day, 593 00:37:15,681 --> 00:37:47,481 if you're starting out on this journey, I think it's really about 594 00:37:47,481 --> 00:38:11,331 committing to that journey, committing to these rules, because 595 00:38:11,331 --> 00:38:48,431 it won't be an easy journey at all for you, Abram, as we discuss 596 00:38:48,431 --> 00:39:14,931 every single week. I think that commitment and really wanting 597 00:39:14,931 --> 00:39:41,431 to do this and feeling completely comfortable with a strategy 598 00:39:41,431 --> 00:40:10,581 that is unpredictable in some ways, yet it's very predictable in 599 00:40:10,581 --> 00:40:34,431 the long run. It's unpredictable from a day-to-day basis. 600 00:40:34,431 --> 00:41:06,231 I think that's probably some of the things I would work on. Allright, 601 00:38:01,415 --> 00:41:06,317 let's move on to Carlo's question. Carlowrites, I would like 602 00:37:33,096 --> 00:38:46,389 to submit a question for Robert Carver. How should a trend 603 00:38:46,389 --> 00:39:46,356 following investor think about position sizing when volatility is 604 00:39:46,356 --> 00:41:06,312 time varying? Okay, so now we're getting into some of the stuff. 605 00:36:59,781 --> 00:37:17,073 Inparticular, I would be very interested in Robert's view on the 606 00:37:17,073 --> 00:37:28,601 following points: static versus dynamic exposure. Some trend 607 00:37:28,601 --> 00:37:41,570 followers argue maximizing the economic benefits of outlier hunting. 608 00:37:41,570 --> 00:37:58,862 (I think he's been listening to Rich too much, actually.) It is 609 00:37:58,862 --> 00:38:11,831 preferable to keep the numbers of contract unchanged throughout 610 00:38:11,831 --> 00:38:27,682 the trade so that only the volatility at entry matters. Others 611 00:38:27,682 --> 00:38:40,651 prefer a continuous volatility targeting framework based on Robert's 612 00:38:40,651 --> 00:38:52,179 backtesting experience. Is there empirical evidence that dynamic 613 00:38:52,179 --> 00:39:05,148 volatility targeting produces better results than a static exposure? 614 00:39:05,148 --> 00:39:19,558 Or is the choice mainly driven by investors objective function? 615 00:39:19,558 --> 00:39:29,645 For example, prioritizing Sharpe ratio optimization versus 616 00:39:29,645 --> 00:39:45,496 maximizing long-term CAGR, as if that is something you can guarantee 617 00:39:45,496 --> 00:40:02,788 just by using static position sizing (I just want to caveat that.) 618 00:40:02,788 --> 00:40:17,198 rebalancing frequency. How often does it make sense to adjust 619 00:40:17,198 --> 00:40:28,726 exposure as volatility changes? Nature of volatility, should 620 00:40:28,726 --> 00:40:38,813 a trend trader treat volatility differently depending 621 00:40:38,813 --> 00:40:53,223 on whether it supports the trend positive volatility or threatens 622 00:40:53,223 --> 00:41:06,192 it’s negative volatility. Thank you so much, from Carlo. 623 00:41:06,881 --> 00:41:10,709 This is a great question. I could almost have written it myself. 624 00:41:10,709 --> 00:41:14,218 So, actually, on the first question about the choice of dynamic 625 00:41:14,218 --> 00:41:17,408 or static volatility position sizing. Actually, let, Let me very 626 00:41:17,408 --> 00:41:21,236 quickly, because I'm sure 90% of the people listening to this know 627 00:41:21,236 --> 00:41:24,426 exactly what we're talking about, but it's probably maybe 10% 628 00:41:24,426 --> 00:41:27,297 who have just started listening. Maybe Abram's just started 629 00:41:27,297 --> 00:41:27,616 listening. 630 00:41:27,641 --> 00:41:28,201 Yep. 631 00:41:28,361 --> 00:41:34,997 So, static exposure is where essentially you put a trade on, and 632 00:41:34,997 --> 00:41:39,974 you size your position according to the volatility that's 633 00:41:39,974 --> 00:41:46,057 current at the time, and then you basically maintain that position 634 00:41:46,057 --> 00:41:53,246 to the same level. Okay, so, if the market gets riskier, you know, 635 00:41:53,246 --> 00:41:59,882 your position stays the same size. If the market gets less risky, 636 00:41:59,882 --> 00:42:05,965 your position stays the same size. You may adjust your position 637 00:42:05,965 --> 00:42:12,048 for other reasons, like if you're using some kind of pyramiding 638 00:42:12,048 --> 00:42:17,578 where you add as things have stronger trends or whatever. Nowthe 639 00:41:33,055 --> 00:41:51,831 alternative is dynamic position sizing, and that's where 640 00:41:51,831 --> 00:42:12,954 you essentially measure the volatility throughout the life of 641 00:42:12,954 --> 00:42:17,648 the trade. So,if the market suddenly gets riskier, you reduce 642 00:41:44,786 --> 00:42:04,861 your position. If the market gets safer, you increase your position. 643 00:42:04,861 --> 00:42:17,636 And this is sort of relates to… 644 00:42:18,131 --> 00:42:20,411 Oh, can I add one thing? 645 00:42:20,411 --> 00:42:20,931 You can. 646 00:42:21,571 --> 00:42:25,091 The dynamic position sizing may not only be influenced by volatility. 647 00:42:25,171 --> 00:42:39,829 I was about to say that, Niels, I was about to say that. Will 648 00:42:39,829 --> 00:42:54,487 you not butt in, please? I'm on a roll here. Come on, give me 649 00:42:54,487 --> 00:42:56,581 some space. So,yeah, what I was going to say was, what I call, 650 00:42:30,547 --> 00:42:34,579 continuous trading which is basically where you constantly evaluate 651 00:42:34,579 --> 00:42:38,163 your positions according to things like volatility, strength 652 00:42:38,163 --> 00:42:43,539 of your trend, but also things like potentially, in fact, it could 653 00:42:43,539 --> 00:42:49,811 be even things like the size of your account as well. So, as you 654 00:42:49,811 --> 00:42:55,635 lose money, you'll be taking money off the table, and so on, and 655 00:42:55,635 --> 00:42:56,531 so forth. 656 00:42:56,581 --> 00:42:58,461 And correlations, Rob, and correlations. 657 00:42:59,261 --> 00:44:29,161 Yeah, potentially, depending on how your risk system works. In 658 00:44:29,161 --> 00:45:32,091 my particular system, correlations wouldn't affect things 659 00:45:32,091 --> 00:46:08,051 to a first order. Butif things get very correlated, then 660 00:43:24,195 --> 00:44:10,501 there is a sort of exogenous risk factor that kicks in. But yes, 661 00:44:10,501 --> 00:44:49,683 if you're running your system at a fixed volatility target, we 662 00:44:49,683 --> 00:45:32,427 always target the same vol every day. Then one thing that would 663 00:45:32,427 --> 00:46:08,047 affect things every day is correlations as well. So, absolutely. 664 00:42:59,261 --> 00:43:55,499 Okay.Now I have tested these two things a few times on my blog. 665 00:43:55,499 --> 00:44:43,703 And one sort of important question is, yes, what, what do you 666 00:44:43,703 --> 00:45:31,907 care about as an investor? Do you care about maximizing your Sharpe 667 00:45:31,907 --> 00:46:08,060 ratio? Do you care about maximizing your geometric returns? 668 00:42:59,261 --> 00:43:53,909 Andthe reason why that's important is that things that have 669 00:43:53,909 --> 00:44:58,493 positive skew may not have as good a Sharpe ratio, but may produce 670 00:44:58,493 --> 00:45:53,141 you a better geometric return. Things that have negative skew produce 671 00:45:53,141 --> 00:46:08,045 the opposite effect. Now,if you're not using volatility targeting, 672 00:43:12,211 --> 00:43:34,411 then your skew profile is much more likely to be positive skew. 673 00:43:34,411 --> 00:43:52,911 And the reason for that is quite straightforward. If you're 674 00:43:52,911 --> 00:44:13,261 holding a position and the thing suddenly leaps in price, and 675 00:44:13,261 --> 00:44:33,611 let's think about cocoa a couple of years ago, because that's 676 00:44:33,611 --> 00:44:53,961 the example that I think still sticks in people's heads. Something 677 00:44:53,961 --> 00:45:12,461 leaps in price, if you're not changing your position according 678 00:45:12,461 --> 00:45:32,811 to volatility, that leap in price, which of course also increases 679 00:45:32,811 --> 00:45:51,311 volatility, you keep your position the same size, and that 680 00:45:51,311 --> 00:46:07,961 means you get this massive outsize positive outlier return. 681 00:42:59,261 --> 00:43:42,161 Andthen that obviously will help improve your geometric return. 682 00:43:42,161 --> 00:44:25,061 Even though, from a Sharpe ratio perspective, it may actually 683 00:44:25,061 --> 00:45:16,541 look worse because your risk is going to be much more variable. 684 00:45:16,541 --> 00:45:59,441 And Sharpe ratio calculations don't like risk that moves around 685 00:45:59,441 --> 00:46:08,021 a lot. Ifyou've got a nice, steady, consistent set of returns, 686 00:43:15,821 --> 00:43:35,693 you're much more likely to have a higher Sharpe ratio. Now it's 687 00:43:35,693 --> 00:43:53,909 basically possible to look at the trade-off between… I won't go 688 00:43:53,909 --> 00:44:17,093 into too much detail because you will have to go and look on my 689 00:44:17,093 --> 00:44:38,621 blog to find the article, but especially you can look at the trade-off 690 00:44:38,621 --> 00:44:56,837 between skew and Sharpe ratio for maximizing your CAGR. So essentially 691 00:44:56,837 --> 00:45:15,053 there should be a certain amount of positive skew that you're 692 00:45:15,053 --> 00:45:36,581 willing to give up in return for higher Sharpe to get the same 693 00:45:36,581 --> 00:45:54,797 CAGR. So, there's sort of almost like a risk\reward trade off, 694 00:45:54,797 --> 00:46:08,045 except that it's skew versus Sharpe ratio trade-off. Andwhat 695 00:43:02,631 --> 00:43:24,536 I found is that basically (and this is not an empirical result by 696 00:43:24,536 --> 00:43:44,756 the way, this is actually theory, so you can't argue with it, 697 00:43:44,756 --> 00:44:04,976 essentially it's not just a quirk of my data), the amount of 698 00:44:04,976 --> 00:44:25,196 Sharpe ratio you should be prepared to give, up for the sort 699 00:44:25,196 --> 00:44:45,416 of improvements in skew that you see from using fixed vol scaling 700 00:44:45,416 --> 00:45:05,636 and not change your position size. The bottom line is it's not 701 00:45:05,636 --> 00:45:22,486 worth doing fixed vol position sizing because the additional skew 702 00:45:22,486 --> 00:45:46,076 you get does not pay you for the loss in Sharpe, which is substantial. 703 00:45:46,076 --> 00:46:07,981 So, you end up with a lower CAGR, basically. And that, you know… 704 00:46:08,301 --> 00:46:09,141 Just to clarify. 705 00:46:09,141 --> 00:46:09,901 Yeah, please go. 706 00:46:09,901 --> 00:46:13,581 Yeah. When you say fixed volume, what you just mean is you 707 00:46:13,581 --> 00:46:16,261 keep the position size static. 708 00:46:16,261 --> 00:50:11,181 Yeah, static position sizing, exactly. Soactually, no, we think 709 00:46:34,901 --> 00:47:12,181 about funds, some CTAs that have extremely good returns, very 710 00:47:12,181 --> 00:47:53,189 good CAGRs, and positive skews, and have this sort of outlier 711 00:47:53,189 --> 00:48:37,925 behavior, and their volatility is often very big. They sort of will 712 00:48:37,925 --> 00:49:18,933 tend to outperform a system that's more like mine, which has 713 00:49:18,933 --> 00:49:56,213 a higher Sharpe ratio, less positive skew, less outliers, because 714 00:49:56,213 --> 00:50:11,125 it's dynamic position sizing. Butbasically, if I apply just a 715 00:46:39,417 --> 00:47:19,113 little bit of extra leverage to my system, I could very easily 716 00:47:19,113 --> 00:47:48,885 beat their geometric returns because my superior Sharpe ratio 717 00:47:48,885 --> 00:48:35,197 more than pays for the loss in positive skew that I have. So that 718 00:48:35,197 --> 00:49:08,277 kind of answers that question. It's quite a long-winded answer, 719 00:49:08,277 --> 00:49:51,281 but it is quite a complex subject. But yeah, go on my blog 720 00:49:51,281 --> 00:50:11,129 and there's an article about it. Thenthere's a question about 721 00:46:34,331 --> 00:47:03,243 rebalancing frequency. Well, essentially this is the standard 722 00:47:03,243 --> 00:47:32,155 question about rebalancing. When you're doing rebalancing, you're 723 00:47:32,155 --> 00:48:11,909 trying to trade off two things: the benefits you'd get from 724 00:48:11,909 --> 00:48:48,049 having your position close to what your optimal position should 725 00:48:48,049 --> 00:49:38,645 be, and the cost of trading to get there. And the worst thing in 726 00:49:38,645 --> 00:50:11,171 the world is if you rebalance the wrong way. So,let's suppose 727 00:46:27,103 --> 00:47:06,857 volatility is sort of jiggling around a bit, but basically falling. 728 00:47:06,857 --> 00:47:46,611 So, all the things being equal, I would increase my position 729 00:47:46,611 --> 00:48:29,979 size. if I was to rebalance, say every day, without fail, then 730 00:48:29,979 --> 00:49:13,347 some days I'd be buying, some days I'd be selling (I'm assuming 731 00:49:13,347 --> 00:49:53,101 everything else is fixed, of course), so I've been carrying a 732 00:49:53,101 --> 00:50:11,171 lot of extra trading costs. So,you know, that would imply that 733 00:46:48,503 --> 00:47:34,563 my rebalancing frequency should be a bit slower because those 734 00:47:34,563 --> 00:48:29,835 extra trading costs are going to more than kill, you know, the 735 00:48:29,835 --> 00:49:20,501 extra benefit I might be getting. But there is a solution 736 00:49:20,501 --> 00:50:11,167 to this conundrum and that's to use either buffering or smoothing. 737 00:46:16,261 --> 00:46:43,018 So,with buffering, essentially you only trade if your 738 00:46:43,018 --> 00:47:15,721 position is further away from the optimal position. And with smoothing, 739 00:47:15,721 --> 00:47:48,424 you basically take something like a moving average of what you 740 00:47:48,424 --> 00:48:30,046 want your position to be. And a lot of trading rules that we use, 741 00:48:30,046 --> 00:48:56,803 obviously incorporate moving averages in them already. So, with 742 00:48:56,803 --> 00:49:26,533 volatility, if you were using something like an extra weighted 743 00:49:26,533 --> 00:49:53,290 moving average of volatility estimates, that would be relatively 744 00:49:53,290 --> 00:50:11,128 smooth and you'd do fewer trades. So,the answer to the question 745 00:46:42,361 --> 00:47:30,211 about rebalancing frequencies, to be honest, I can't tell you because 746 00:47:30,211 --> 00:48:35,461 it's going to depend on costs. But as I said, if you use these other 747 00:48:35,461 --> 00:49:14,611 strategies, then you can do rebalancing more frequently without 748 00:49:14,611 --> 00:50:02,461 actually incurring any trading costs. So, you're going to get the 749 00:50:02,461 --> 00:50:11,161 win there. Now,the last question actually is the most intriguing 750 00:46:52,090 --> 00:47:39,862 because I don't actually have an answer for it. And that's about 751 00:47:39,862 --> 00:48:23,653 using an asymmetric volatility measure. So, if we think about cocoa 752 00:48:23,653 --> 00:49:07,444 as an example, then that was positive volatility, right? The price 753 00:49:07,444 --> 00:49:51,235 went up, we were long, Hooray! And maybe in those circumstances 754 00:49:51,235 --> 00:50:11,140 we shouldn't reduce our position. Now,my concern with this 755 00:46:28,246 --> 00:46:59,407 is that returns are a coin flip pretty much. So, if you're a 756 00:46:59,407 --> 00:47:30,568 really good trader, you might make money for 51% of days and lose 757 00:47:30,568 --> 00:48:01,729 money 49% of days. So, if you've got a massive risk on because 758 00:48:01,729 --> 00:48:30,493 the thing's just gone up in a straight line and that's wonderful, 759 00:48:30,493 --> 00:48:56,860 and you just hold your position the same size, you're taking 760 00:48:56,860 --> 00:49:28,021 this huge amount of upside of risk. Your upside risk and your downside 761 00:49:28,021 --> 00:49:56,785 risk are the same. Okay, you're just a coin flip away from 762 00:49:56,785 --> 00:50:11,167 those massive profits becoming massive losses. Sooverall, I'd say, 763 00:46:32,745 --> 00:47:22,197 no. I'd say you should treat both kinds of volatility the same. 764 00:47:22,197 --> 00:48:07,528 That's not to say, though, that assets do have different behavior 765 00:48:07,528 --> 00:48:48,738 on the upside and the downside. And there are interesting 766 00:48:48,738 --> 00:49:34,069 things we can do to model that. But this specific question, 767 00:49:34,069 --> 00:50:11,158 I suggest that you just treat risk as symmetrical. 768 00:50:12,572 --> 00:50:29,127 Well, that's definitely an option for you to include that in 769 00:50:29,127 --> 00:50:45,682 one of your next books, whether there is a difference in 770 00:50:45,682 --> 00:50:48,692 that subject. Anyways,let's move on because we've still got a 771 00:50:19,961 --> 00:50:30,634 couple of questions before we get to the big secret topic of today. 772 00:50:30,634 --> 00:50:40,486 A question from Dario. Hi, Mr. Carver. I hope you're well. Thank 773 00:50:40,486 --> 00:50:48,696 you so much for your contributions to the investment community. 774 00:50:12,572 --> 00:50:22,427 I'mcurious about sentiment indicators used in low frequency 775 00:50:22,427 --> 00:50:34,472 algorithms. I wonder if you have ever tried trading off sentiment 776 00:50:34,472 --> 00:50:45,422 indicators? Are your skew signals a proxy for sentiment? Thank 777 00:50:45,422 --> 00:50:48,707 you, from Dario. 778 00:50:51,272 --> 00:51:16,442 I must say, I'm slightly uncomfortable with all these people 779 00:51:16,442 --> 00:51:49,163 calling me Mr. Carver. It's a bit formal for me, but anyway. Okay, 780 00:51:49,163 --> 00:52:14,333 so, sentiment is interesting because I don't actually use any 781 00:52:14,333 --> 00:52:39,503 sort of sentiment indicators myself. So, my skew signal, which 782 00:52:39,503 --> 00:53:02,156 he asked specifically about, that's something that's looking at 783 00:53:02,156 --> 00:53:12,224 historic levels of skew. Now,you could argue that as a sentiment 784 00:51:01,336 --> 00:51:15,174 indicator. You could argue that if an asset has recently seen 785 00:51:15,174 --> 00:51:29,012 very violent falls in price, that suggests that most people would 786 00:51:29,012 --> 00:51:42,850 have a negative sentiment towards that asset. And in the case 787 00:51:42,850 --> 00:51:57,946 of my trading system, I would actually buy that asset because something 788 00:51:57,946 --> 00:52:11,784 that's got very heavy negative skew would tend to be underpriced 789 00:52:11,784 --> 00:52:25,622 because people don't like it. So, actually, in that specific example. 790 00:52:25,622 --> 00:52:36,944 I'm a buyer of negative sentiment, But generally speaking, 791 00:52:36,944 --> 00:52:52,040 partly because of resources, I tend to stay away from, say, weirder 792 00:52:52,040 --> 00:53:05,878 data, things like sentiment scores. Also, some of them tend to 793 00:53:05,878 --> 00:53:12,168 be, effectively, in sample fitted. So,when I used to work for 794 00:51:08,576 --> 00:51:38,240 AHL, we had a classic thing where sell side traders would come 795 00:51:38,240 --> 00:52:05,432 in and say, look, we've got this amazing model that predicts 796 00:52:05,432 --> 00:52:42,512 this, do you want to buy it, or trade it through us and give us 797 00:52:42,512 --> 00:53:09,704 a commission, or whatever. And you'd say, how did you construct 798 00:53:09,704 --> 00:53:12,176 it? Andafter a lot of plotting ,and prodding, and pulling, 799 00:51:19,462 --> 00:51:50,471 and asking your questions, and digging into details, you find out 800 00:51:50,471 --> 00:52:18,661 essentially it's massively in sample fitted and therefore of course 801 00:52:18,661 --> 00:52:46,851 it looks very good. And we, generally speaking, would prefer 802 00:52:46,851 --> 00:53:12,222 to do that job ourselves, and do it properly. So,I'd be wary of 803 00:50:57,917 --> 00:51:13,865 kind of buying any or sort of random, third party sentiment indicators, 804 00:51:13,865 --> 00:51:29,813 but there are things that kind of indicate the sentiment of the 805 00:51:29,813 --> 00:51:44,432 market. So skew is one. There's sort of various options ratios 806 00:51:44,432 --> 00:52:04,367 as well you can use. You could just look at the level of the VIX 807 00:52:04,367 --> 00:52:21,644 and say, well, that's sort of a proxy for how scared people are. 808 00:52:21,644 --> 00:52:37,592 That's a measure of sentiment. So, all of these are good things 809 00:52:37,592 --> 00:52:50,882 that are potentially useful for predicting prices. I just tend 810 00:52:50,882 --> 00:53:08,159 not to use them myself. But that doesn't mean that it's not a 811 00:53:08,159 --> 00:53:12,146 valid research area. 812 00:53:12,902 --> 00:53:16,362 Sure, fine. Final question today is from Andreas, actually I 813 00:53:16,362 --> 00:53:20,860 know who Andreas is. So, he has two questions for you Rob, and 814 00:53:20,860 --> 00:53:25,358 he does call you Rob, by the way. In Rob's research, the relationship 815 00:53:25,358 --> 00:53:28,818 between ATR and standard deviation, he concludes with an empirical 816 00:53:28,818 --> 00:53:31,932 and theoretical solution. Question, A, is there a mathematical 817 00:53:31,932 --> 00:53:35,392 logical relationship between daily and weekly ATR? And, B, if 818 00:53:35,392 --> 00:53:40,236 a model is trading once a week only, is it advisable to use daily 819 00:53:40,236 --> 00:53:44,042 data or weekly data with respective to ATR? And then there's 820 00:53:44,042 --> 00:53:47,848 a follow up question, but let's do one at a time. 821 00:53:47,862 --> 00:54:09,521 Yeah, so, ATR (for those who aren't familiar) is Average True 822 00:54:09,521 --> 00:54:29,211 Range. So, the true range essentially is the difference in 823 00:54:29,211 --> 00:54:56,777 the highest and lowest prices you see in a time period like a day. 824 00:54:56,777 --> 00:55:22,374 So, it's sort of a bit like standard deviation in the sense that 825 00:55:22,374 --> 00:55:42,064 it's designed to measure how much market smoothing… Rather than 826 00:55:42,064 --> 00:56:01,754 just looking at single price per time interval, what technical 827 00:56:01,754 --> 00:56:29,320 analysts call a bar, it looks at the size of the bar itself as 828 00:56:29,320 --> 00:56:31,289 well. Now,essentially, it's not possible to work out… So, you 829 00:54:18,702 --> 00:54:49,542 could look at the imperial relationships in daily and weekly 830 00:54:49,542 --> 00:55:17,298 ATRs, but it depends on essential autocorrelation. It depends 831 00:55:17,298 --> 00:55:57,390 on how the price return in one period influences a return in the 832 00:55:57,390 --> 00:56:22,062 following period. And this actually affects standard deviation 833 00:56:22,062 --> 00:56:31,314 estimates as well. So,for example, returns tend to cluster, 834 00:53:57,319 --> 00:54:13,531 so good returns tend to be followed by other good returns, then 835 00:54:13,531 --> 00:54:25,690 generally speaking, if you compare, say, your daily standard 836 00:54:25,690 --> 00:54:36,498 deviation, your annual standard deviation, you'll find that 837 00:54:36,498 --> 00:54:51,359 there isn't the kind of theoretical square root of time relationship 838 00:54:51,359 --> 00:55:07,571 between the two. You'll find that… So, let's take an extreme example. 839 00:55:07,571 --> 00:55:27,836 Suppose prices go up by 1% and down by 1% every day of the year. 840 00:55:27,836 --> 00:55:46,750 If it's a leap year, there's an even number of days in the year, 841 00:55:46,750 --> 00:56:04,313 so the return for the year will be zero. So, if that happened 842 00:56:04,313 --> 00:56:16,472 forever, then the standard deviation of annual returns would 843 00:56:16,472 --> 00:56:31,333 be zero. But the standard deviation of daily returns is 1%. 844 00:53:47,862 --> 00:54:37,813 So,that simple example where you've got very strong negative autocorrelation, 845 00:54:37,813 --> 00:55:27,764 the daily standard deviation is much higher than the annual standard 846 00:55:27,764 --> 00:56:13,174 deviation. If you had positive autocorrelation, then it would be 847 00:56:13,174 --> 00:56:31,338 the way around, basically. Andthis is quite interesting for 848 00:54:02,076 --> 00:54:30,504 people who do things like… There's a paper by Andrew Lowe where 849 00:54:30,504 --> 00:54:54,194 he looks at hedge fund performance and Sharpe ratios, and 850 00:54:54,194 --> 00:55:24,991 concludes that he really wants to try and get sort of monthly, weekly, 851 00:55:24,991 --> 00:55:51,050 or daily data rather than just looking at annual because annual 852 00:55:51,050 --> 00:56:14,740 returns hide a lot of fun autocorrelation properties. But that's 853 00:56:14,740 --> 00:56:31,323 another story. So yeah, it's not really… So,I think off the top 854 00:53:59,820 --> 00:54:21,743 of my head, the weekly to daily ATR relationship should behave 855 00:54:21,743 --> 00:54:41,673 the same as standard deviation. If you assume zero autocorrelation 856 00:54:41,673 --> 00:54:57,617 and probably some of the distributional assumptions, then 857 00:54:57,617 --> 00:55:21,533 it should follow the square root of time rule. And that means 858 00:55:21,533 --> 00:55:43,456 that if markets are trading on weekdays only, then there should 859 00:55:43,456 --> 00:56:09,365 be a sort of multiple of about square root of five between those 860 00:56:09,365 --> 00:56:31,288 two values, in theory. But yeah, it depends on the autocorrelation. 861 00:56:31,842 --> 00:56:35,235 And would you use weekly data if you only trade once a week? 862 00:56:35,242 --> 00:56:50,127 I mean, yeah, there's no reason not to. It was like the earlier 863 00:56:50,127 --> 00:57:02,722 discussion when we were saying, well, why use tick data if 864 00:57:02,722 --> 00:57:14,172 you're trading daily? What extra information are you going to 865 00:57:14,172 --> 00:57:15,317 get? Now,having said that… Sorry, the question was about estimating 866 00:56:40,882 --> 00:56:48,214 the ATR. I would say yes, you always get a more accurate estimation 867 00:56:48,214 --> 00:56:53,854 of volatility if you use data that's more frequent, definitely. 868 00:56:53,854 --> 00:57:00,622 So yeah, I would use daily price changes to calculate my ATRs. 869 00:57:00,622 --> 00:57:07,390 But in terms of actual general backtesting of my system, then yeah, 870 00:57:07,390 --> 00:57:14,158 sure, weekly for weekly, daily to daily. There's no reason to do 871 00:57:14,158 --> 00:57:15,286 it faster. 872 00:57:15,752 --> 00:57:36,632 Yeah, that's important. Allright, final second part of the 873 00:57:18,734 --> 00:57:22,994 question. Has Rob done any research on using intraday stops 874 00:57:22,994 --> 00:57:28,532 versus using market orders at the next day open if the hard stop 875 00:57:28,532 --> 00:57:32,792 has been reached? How much additional room e.g. expressed in 876 00:57:32,792 --> 00:57:36,626 ATRs is advisable to add to an intraday stop? 877 00:57:37,672 --> 00:57:46,772 I mean, I don't use stop losses when I'm trading and I don't 878 00:57:46,772 --> 00:57:55,872 use ATRs either. So, I'm not sure how qualified I am to answer 879 00:57:55,872 --> 00:58:05,672 this question. The one thing I will say is that there is a nice 880 00:58:05,672 --> 00:58:14,072 paper by… I can't remember who it's by. It's inevitably going to 881 00:58:14,072 --> 00:58:23,172 be someone who works at AQR, I'll say that much. I don't think 882 00:58:23,172 --> 00:58:32,972 it's Anty and I don't think it's Lars. I think it might be Frank 883 00:58:32,972 --> 00:58:39,972 anyway. But the paper basically discusses a very neat way 884 00:58:39,972 --> 00:58:45,572 of combining slow momentum with shorter-term mean reversion 885 00:58:45,572 --> 00:58:53,972 is to do something really simple, which is, when you come to 886 00:58:53,972 --> 00:59:01,672 execute your trades the day after the previous day's close (as 887 00:59:01,672 --> 00:59:08,672 we've been discussing), only execute those that have moved in 888 00:59:08,672 --> 00:59:16,372 your favor. So, if you're buying, only execute if the price 889 00:59:16,372 --> 00:59:23,372 has dropped in the last closing price and vice versa. So,that's 890 00:57:41,198 --> 00:58:02,354 a very, a very, quite a cheap way of getting potentially better 891 00:58:02,354 --> 00:58:19,984 execution costs by essentially combining your slow system with a 892 00:58:19,984 --> 00:58:35,851 fast system. Because normally the problem, as we've discussed, 893 00:58:35,851 --> 00:58:57,007 with faster systems that cost too much in trading costs. This way 894 00:58:57,007 --> 00:59:18,163 you're going to do that trade anyway. It's just which level you 895 00:59:18,163 --> 00:59:23,452 do it at. So,the key issue there is whether that delay of one, 896 00:57:51,686 --> 00:58:10,796 two, maybe more days… We know that a delay of one day is not going 897 00:58:10,796 --> 00:58:24,810 to affect us too much. We discussed that already. But whether 898 00:58:24,810 --> 00:58:37,550 delaying by potentially more than one day, because you're waiting 899 00:58:37,550 --> 00:58:54,112 for the price to reach a nice level, whether that's an issue or 900 00:58:54,112 --> 00:59:09,400 not. And that's something I need to test because that's in my 901 00:59:09,400 --> 00:59:23,414 back book of things I need to research and implement, definitely. 902 00:57:37,672 --> 00:58:08,285 I'venot answered the question at all because I feel completely 903 00:58:08,285 --> 00:58:44,464 unqualified to. It's just that was a random idea that came out of 904 00:58:44,464 --> 00:59:23,426 my head that I think is sort of related to what he's talking about. 905 00:59:29,932 --> 00:59:38,501 Let's move on to the very important topic, I think, because 906 00:59:38,501 --> 00:59:47,849 yesterday, I think it was, there was a vote in California, not 907 00:59:47,849 --> 00:59:56,418 about election districts or anything like that. It was from the 908 00:59:56,418 --> 01:00:04,987 board of administration of US pension fund CalPERS and they voted 909 01:00:04,987 --> 01:00:13,556 in favor of adopting the total portfolio approach, TPA, which is 910 01:00:13,556 --> 01:00:19,788 going to replace their existing strategic asset allocation, 911 01:00:19,788 --> 01:00:29,915 the SAA, model next year, July 1, 2026. And they also said that 912 01:00:29,915 --> 01:00:38,484 they are going to change their model reference portfolio to 75% 913 01:00:38,484 --> 01:00:48,611 equities, 25% bonds. And they also said (and this is a quote because 914 01:00:48,611 --> 01:00:58,738 you and I were not entirely sure whether this was true or not, 915 01:00:58,738 --> 01:01:08,865 but it is a quote), they say, “they are the first pension fund 916 01:01:08,865 --> 01:01:18,213 in the United States to adopt TPA.” And David Miller, the investment 917 01:01:18,213 --> 01:01:26,003 committee chairman said this will give CalPERS’ staff the edge 918 01:01:26,003 --> 01:01:31,456 they need to make sound investment decisions. Now,in the 919 00:59:34,381 --> 00:59:53,660 article that I found, which is not the FT article that you sent, 920 00:59:53,660 --> 01:00:14,422 Rob, it also goes on to say that under the TPA, the focus will 921 01:00:14,422 --> 01:00:29,252 be on which investments can best contribute to the performance 922 01:00:29,252 --> 01:00:42,599 of the entire CalPERS portfolio as opposed to achieving 923 01:00:42,599 --> 01:00:54,463 individual asset class targets that were periodically reviewed. 924 01:00:54,463 --> 01:01:15,225 Now, I'm going to shut up now pretty much because I want you to 925 01:01:15,225 --> 01:01:31,538 take us through what you found to be the interesting part. Butas 926 00:59:35,220 --> 01:00:09,592 I said last week, when I spoke with Katy, because of that sentence, 927 01:00:09,592 --> 01:00:41,320 that each component will be judged on what they add in terms 928 01:00:41,320 --> 01:01:10,404 of value to the portfolio, personally, I think that makes it 929 01:01:10,404 --> 01:01:31,556 very, very interesting from a trend following perspective. 930 01:01:32,302 --> 01:01:45,302 I think it does. Yeah. Imean,although cynically I wonder 931 01:01:34,990 --> 01:01:41,262 whether this is just going to be used an excuse to jam more private 932 01:01:41,262 --> 01:01:45,294 equity, you know, or private debt, or other private… 933 01:01:45,302 --> 01:01:45,902 Anything private. 934 01:01:45,902 --> 01:04:03,655 That's because people love the private stuff at the moment, don't 935 01:04:03,655 --> 01:06:33,931 they? But now, let's be upbeat and positive and assume it's going 936 01:06:33,931 --> 01:08:14,115 to be a good thing for trend following. Now,I have to say I was 937 01:02:37,170 --> 01:04:12,382 very intrigued by this subject for a number of reasons. One is that 938 01:04:12,382 --> 01:05:40,270 portfolio optimization is sort of my hobby, I suppose it's one of 939 01:05:40,270 --> 01:07:15,482 the big research area I keep coming back to and one I've done 940 01:07:15,482 --> 01:08:14,074 a lot of spending time and thinking about. So,the sort of basic 941 01:02:01,812 --> 01:02:33,632 maths of the Markovits, sort of so-Called Modern Portfolio Theory, 942 01:02:33,632 --> 01:03:11,816 modern even though it was invented in the 1950s, and has underlied 943 01:03:11,816 --> 01:03:50,000 pretty much everything ever since and is what the kind of asset 944 01:03:50,000 --> 01:04:21,820 allocation industry use as well. So, though I've never worked 945 01:04:21,820 --> 01:04:56,822 as an asset allocator, obviously I've had to deal with them 946 01:04:56,822 --> 01:05:38,188 as a potential supplier of alpha, as one might put it, and also 947 01:05:38,188 --> 01:06:16,372 because Man Group had a number of multi manager strands within it. 948 01:06:16,372 --> 01:06:51,374 I've also worked with the quants working as multi manager organizations, 949 01:06:51,374 --> 01:07:26,376 helping them with things like analyzing returns. So, I have some 950 01:07:26,376 --> 01:07:58,196 understanding of the way that asset allocators think. But let's 951 01:07:58,196 --> 01:08:14,106 sort of briefly discuss this. So,SAA or strategic asset allocation, 952 01:02:27,494 --> 01:03:57,610 is kind of the way that people have done asset allocation for almost, 953 01:03:57,610 --> 01:05:13,862 I would say, my entire lifetime, pretty much with some tweaks. 954 01:05:13,862 --> 01:06:43,978 And it does grow out of the Markovitz model. And the basic idea 955 01:06:43,978 --> 01:08:07,162 is this, it's simplest to think about it in an equity bond 956 01:08:07,162 --> 01:08:14,094 setup. So,you think about your expectations for forward returns, 957 01:02:38,840 --> 01:03:37,660 risk and correlations between the two asset classes that you're 958 01:03:37,660 --> 01:04:42,362 considering, and then you set some target return. And that's for 959 01:04:42,362 --> 01:05:52,946 example, if you're a pension fund, you know that's going to be 960 01:05:52,946 --> 01:07:03,530 based on something like the average age of your members and you 961 01:07:03,530 --> 01:08:14,114 know how much money you need to sort of pay them out. Youknow, 962 01:02:03,548 --> 01:03:40,601 a recent innovation has been these things called target date funds, 963 01:03:40,601 --> 01:05:08,831 where you dynamically change your allocation as people age effectively. 964 01:05:08,831 --> 01:07:03,530 And you know, in simple terms, that will put more into bonds and 965 01:07:03,530 --> 01:08:14,114 less into equities, which kind of makes sense. Andthen you build 966 01:02:08,086 --> 01:02:58,000 this thing called the efficient frontier, which is the 967 01:02:58,000 --> 01:04:10,098 sort of the best portfolio of these two assets. You pick a point 968 01:04:10,098 --> 01:05:22,196 that has the highest Sharpe ratio and then, if you can use leverage, 969 01:05:22,196 --> 01:06:17,656 you basically construct a tangent. If you can't use leverage, 970 01:06:17,656 --> 01:07:18,662 you pick some point along the efficient frontier where you get, 971 01:07:18,662 --> 01:08:14,122 essentially, the lowest risk (that still meets your hurdle) return. 972 01:01:45,902 --> 01:02:18,495 So,what that means in practice is if you're a retirement 973 01:02:18,495 --> 01:02:54,051 fund, but your members are quite young and you want to achieve 974 01:02:54,051 --> 01:03:35,533 a certain amount, you can sort of have a higher return to try and 975 01:03:35,533 --> 01:04:11,089 meet your obligations in the future. And that will mean a higher 976 01:04:11,089 --> 01:04:43,682 mix to equities. And obviously, if you've got a lot of 977 01:04:43,682 --> 01:05:19,238 old retirees, and this is particularly true of the UK… in the 978 01:05:19,238 --> 01:05:48,868 UK we've got a lot of so-called defined benefit pension 979 01:05:48,868 --> 01:06:24,424 funds where the investment risk is all on the manager and not 980 01:06:24,424 --> 01:07:02,943 the person, the retiree. And that means that they have a lot of 981 01:07:02,943 --> 01:07:32,573 money in bonds. And that's been particularly good for long-dated 982 01:07:32,573 --> 01:08:05,166 UK bonds, actually. That does weird things to the yield curve. 983 01:08:05,166 --> 01:08:14,055 That's another story. Anyway,that's just the starting 984 01:02:10,787 --> 01:03:05,534 point. And then you’ve got your SAA or strategic asset allocation, 985 01:03:05,534 --> 01:04:10,235 and then, of course, you can add to that other things. So, they 986 01:04:10,235 --> 01:04:55,028 use this expression ‘sleeves’, which I've never really understood 987 01:04:55,028 --> 01:05:54,752 but you can have sleeves for hedge funds, you know, for equity 988 01:05:54,752 --> 01:06:39,545 market neutral hedge funds specifically, perhaps. You can have 989 01:06:39,545 --> 01:07:34,292 them for private credit and again for private equity. And of 990 01:07:34,292 --> 01:08:14,108 course you could have a sleeve for CTAs. Andthen the idea is that 991 01:02:17,378 --> 01:03:30,822 you then sort of go down to the next level and you say, right, 992 01:03:30,822 --> 01:04:39,020 we need to have exposure to this asset class. And then we maybe 993 01:04:39,020 --> 01:05:36,726 make some determination about how much is passive, how much is 994 01:05:36,726 --> 01:06:34,432 active. Maybe we have some country weightings and then so on, 995 01:06:34,432 --> 01:07:42,630 and so forth. So, the key points here are that it's a very 996 01:07:42,630 --> 01:08:14,122 kind of structured top-down process, basically. Iliketo think 997 01:02:13,630 --> 01:03:57,610 of it as a series of boxes. So you start off by saying, well, I've 998 01:03:57,610 --> 01:05:41,590 got this big box and in it I need to put asset classes in different 999 01:05:41,590 --> 01:07:11,706 sizes, and then I open up those boxes, and I put other stuff 1000 01:07:11,706 --> 01:08:14,094 inside those boxes, and so on, and so forth. Andimportantly, no 1001 01:01:59,603 --> 01:02:58,974 one takes into view the fact that if, for example, let's say I 1002 01:02:58,974 --> 01:03:53,778 happen to put into one of my boxes something that's highly correlated 1003 01:03:53,778 --> 01:04:48,582 to something in another box, that does not enter into the decision 1004 01:04:48,582 --> 01:05:38,819 process because no, no, no, we're just putting stuff in boxes 1005 01:05:38,819 --> 01:06:24,489 now. We're not considering the total portfolio, the holistic portfolio. 1006 01:06:24,489 --> 01:07:14,726 And that might mean, for example, that, well, let's take what 1007 01:07:14,726 --> 01:08:14,097 I think is a pretty made-up example. Let's just think of it anyway. 1008 01:01:45,902 --> 01:02:27,768 Let'ssuppose you have a CTA that has amazing performance but 1009 01:02:27,768 --> 01:03:09,634 for some reason has a very high correlation with the Danish 1010 01:03:09,634 --> 01:03:43,888 equity market. (This is a completely made-up example.) And 1011 01:03:43,888 --> 01:04:21,948 let's also suppose that, separately, you're the guy that manages 1012 01:04:21,948 --> 01:04:56,202 your equity portfolio, really likes Danish equities. Well, now 1013 01:04:56,202 --> 01:05:41,874 you've got a problem because you have a big exposure to Danish 1014 01:05:41,874 --> 01:06:23,740 equities in two parts of your portfolio. And maybe somewhere there's 1015 01:06:23,740 --> 01:07:13,218 a risk management team that will point this out to you. But in 1016 01:07:13,218 --> 01:07:47,472 terms of the portfolio construction process, this sort of 1017 01:07:47,472 --> 01:08:14,114 weird correlation hasn't been taken into account. So,my understanding 1018 01:01:56,117 --> 01:02:33,572 of total portfolio approach is that, rather than doing this kind 1019 01:02:33,572 --> 01:03:14,432 of top-down bits and pieces, blah, blah, blah, that we look at 1020 01:03:14,432 --> 01:03:55,292 the thing as a whole… And then we'd potentially make a decision 1021 01:03:55,292 --> 01:04:32,747 along lines of saying, well, we've got this amazing CTA manager, 1022 01:04:32,747 --> 01:05:10,202 but he's got this massive exposure to Danish equities and we've 1023 01:05:10,202 --> 01:05:51,062 already got a big exposure to Danish equities. Well, does it make 1024 01:05:51,062 --> 01:06:31,922 sense to add them to the portfolio? Does it make sense to 1025 01:06:31,922 --> 01:07:09,377 dial down the long-only Danish equities and then add these guys 1026 01:07:09,377 --> 01:07:53,642 in, or does it make sense to take out the Danish equities long-only 1027 01:07:53,642 --> 01:08:14,072 and just put the CTA in? So,you make those decisions on a 1028 01:03:13,563 --> 01:05:56,362 more holistic basis. Now, I have to say that, to an extent, this 1029 01:05:56,362 --> 01:08:14,115 whole…. Can we call it an industry yet? I don't know. 1030 01:08:14,362 --> 01:08:14,682 Which part? 1031 01:08:14,682 --> 01:08:16,290 Well, the TPA in industry, kyounow. 1032 01:08:16,290 --> 01:08:16,562 Well, they are alike. 1033 01:08:16,562 --> 01:08:18,638 There are people who are going to do this for a job. 1034 01:08:18,882 --> 01:08:25,458 They quote like five or six sovereign wealth funds or big pension 1035 01:08:25,458 --> 01:08:31,486 funds worldwide. I'm not so sure that qualifies it just yet. 1036 01:08:31,501 --> 01:08:34,999 Let's call it a cult then, instead, because it's like maybe 1037 01:08:34,999 --> 01:08:38,497 a few hundred people doing this globally. That's about the size 1038 01:08:38,497 --> 01:08:39,451 of a cult. 1039 01:08:39,461 --> 01:08:40,621 Or a group, we could just call them a group. 1040 01:08:40,621 --> 01:10:02,593 A group. I don't know about cult, maybe, has connotations, do 1041 01:10:02,593 --> 01:10:17,497 you think? Anyway,one of the problems I found with this group 1042 01:08:49,841 --> 01:09:00,905 is that it does seem to be something that's been invented by 1043 01:09:00,905 --> 01:09:10,125 some management consultants and it's very vague and there's a 1044 01:09:10,125 --> 01:09:20,267 lot of very nice charts and pictures and expressions like… (I'm 1045 01:09:20,267 --> 01:09:31,331 going to read this out.) “TPA shifts the focus from rigid allocations 1046 01:09:31,331 --> 01:09:40,551 to unified strategy where decisions are guided by the Fund's 1047 01:09:40,551 --> 01:09:48,849 overarching goals, drawing on ideas from the entire investment 1048 01:09:48,849 --> 01:09:56,225 portfolio, inducing the flexibility to adapt to emerging 1049 01:09:56,225 --> 01:10:00,835 priorities such as sustainability, intergenerational 1050 01:10:00,835 --> 01:10:07,289 equity, evolving economic conditions. The approach enhances 1051 01:10:07,289 --> 01:10:12,821 governance, fosters collaboration, unlocks new opportunities 1052 01:10:12,821 --> 01:10:17,431 and supports multidimensional 3D investing.” Now,that sentence 1053 01:08:46,102 --> 01:09:04,372 was definitely written by a management consultant and not by 1054 01:09:04,372 --> 01:09:24,469 a quant, because a quant would say something like maximize utility 1055 01:09:24,469 --> 01:09:40,912 function, where you input the correlation, the standard deviation, 1056 01:09:40,912 --> 01:09:59,182 and the expected vector of mean returns, plus your constraints, 1057 01:09:59,182 --> 01:10:15,625 and with that I've just described essentially the Markovitz 1058 01:10:15,625 --> 01:10:17,452 optimization. Soit's a bit hard for me to kind of make judgments 1059 01:08:51,793 --> 01:09:01,103 on this, generally speaking, because there's no real precise mathematical 1060 01:09:01,103 --> 01:09:14,137 definition of what it is. And, you know, I really feel like I need 1061 01:09:14,137 --> 01:09:27,171 that to hang on to. But, as a general rule, clearly, I think it 1062 01:09:27,171 --> 01:09:39,274 makes a lot more sense to think about your portfolio in a holistic 1063 01:09:39,274 --> 01:09:48,584 sense, rather than just putting things in little boxes. Because 1064 01:09:48,584 --> 01:10:01,618 I think another point I would make is that one of the issues the 1065 01:10:01,618 --> 01:10:15,583 CTA industry has is which box do we go in? Are we in the hedge 1066 01:10:15,583 --> 01:10:17,445 fund box? 1067 01:10:18,652 --> 01:10:31,232 Because we're quite different from most hedge funds, we've got 1068 01:10:31,232 --> 01:10:46,328 quite a different correlation. Are we in some kind of nebulous tail 1069 01:10:46,328 --> 01:10:57,650 protection box? Well, that's tricky because unlike proper tail 1070 01:10:57,650 --> 01:11:10,230 protection funds, we can't guarantee that we'll make you money 1071 01:11:10,230 --> 01:11:26,584 in a down market. We hope we will, but there's no guarantee of 1072 01:11:26,584 --> 01:11:41,680 it. On the other hand, we've got a positive Sharpe ratio, which 1073 01:11:41,680 --> 01:11:56,776 those guys don't have, so we don't fit in that box either. Somaybe, 1074 01:10:21,670 --> 01:10:38,269 just maybe, this move away from a reductionist approach of putting 1075 01:10:38,269 --> 01:10:53,359 everything into little boxes will also be good because people 1076 01:10:53,359 --> 01:11:12,976 will be able to say, well, I've got my portfolio here. And if 1077 01:11:12,976 --> 01:11:31,084 I add this weird thing here, it does improve the overall setup 1078 01:11:31,084 --> 01:11:49,192 (as long as there's not any too many Danish equities in there, 1079 01:11:49,192 --> 01:11:56,737 of course), and that's great. So,that's kind of where I am. I'm 1080 01:10:29,404 --> 01:10:44,188 hopeful. I'm a little bit skeptical because it does sound a 1081 01:10:44,188 --> 01:10:56,284 little bit management consultancy. I'm also, as a quant, 1082 01:10:56,284 --> 01:11:09,724 struggling with the fact there's not really any formal definition 1083 01:11:09,724 --> 01:11:31,228 of how you do this. And there does seem to be a lot of hand waving 1084 01:11:31,228 --> 01:11:48,700 going on. But that's my take on it. But I think you're right, 1085 01:11:48,700 --> 01:11:56,782 this could be a game changer. Imean,if CalPERS are doing it, 1086 01:10:26,828 --> 01:10:39,676 they're big players. And if, for example, someone like ADIA starts 1087 01:10:39,676 --> 01:10:54,860 looking at it, the quant team at ADIA, I think they employ some 1088 01:10:54,860 --> 01:11:10,044 like 20% of all the quants of the world now. You know, there's 1089 01:11:10,044 --> 01:11:24,060 some very famous people there. So, if they start doing it, then 1090 01:11:24,060 --> 01:11:39,244 I think that really will be a game changer and maybe they'll actually 1091 01:11:39,244 --> 01:11:54,428 write some papers and explain to us what it is, because I'm still 1092 01:11:54,428 --> 01:11:56,764 struggling slightly. 1093 01:11:57,262 --> 01:12:13,362 Yeah, I think you make a lot of great points. And I was thinking, 1094 01:12:13,362 --> 01:12:26,012 if it wasn't a management consultant that wrote that text you 1095 01:12:26,012 --> 01:12:39,812 read, it could have been chatGPT. And the reason I say that 1096 01:12:39,812 --> 01:12:54,762 is you also sent me an article, and I have no idea what 1097 01:12:54,762 --> 01:13:08,562 you were thinking, you sent me an article in French expecting me 1098 01:13:08,562 --> 01:13:20,062 to read about the total portfolio approach in French. And, 1099 01:13:20,062 --> 01:13:33,862 to be very open about it, I'm not very good at French. Anyways,so 1100 01:11:59,508 --> 01:12:15,230 I had to resort to ChatGPT just to make some sense of it. And 1101 01:12:15,230 --> 01:12:32,075 what it did, I just asked for a summary and when it got to the 1102 01:12:32,075 --> 01:12:44,428 TPA it wrote the following. (And that's why I'm thinking that 1103 01:12:44,428 --> 01:12:55,658 it could be ChatGPT.) “One portfolio, one goal, maximize total 1104 01:12:55,658 --> 01:13:06,888 return within a defined risk budget. Asset classes don't matter, 1105 01:13:06,888 --> 01:13:18,118 only contribution to total outcome does. CIO lead, dynamic collaboration, 1106 01:13:18,118 --> 01:13:27,102 performance equal fund objective, not asset class alpha, 1107 01:13:27,102 --> 01:13:33,840 this is portfolio construction without walls.” Andthen I asked, 1108 01:12:04,990 --> 01:12:28,174 okay, so why does this work? And it says,”It empowers teams to 1109 01:12:28,174 --> 01:12:53,290 act on the best opportunities, not just stay in their lane. It frees 1110 01:12:53,290 --> 01:13:10,678 capital for macro, trend and uncorrelated strategies and puts 1111 01:13:10,678 --> 01:13:31,930 risk at the center where it belongs. It favors adaptability over 1112 01:13:31,930 --> 01:13:33,862 prediction.” 1113 01:13:33,872 --> 01:13:36,194 Yeah, I don't know. It's becoming increasingly difficult to 1114 01:13:36,194 --> 01:13:39,032 tell the difference between AI and non AI related content, especially 1115 01:13:39,032 --> 01:13:41,870 if it comes from a management consultant, I have to say. 1116 01:13:41,872 --> 01:13:48,412 It even went on to say, “Here's the TTU, the Top Traders 1117 01:13:48,412 --> 01:13:56,042 Unplugged take. Okay, it says SAA is a map, TPA is GPS, and in 1118 01:13:56,042 --> 01:14:01,492 this environment you need to navigate, not just rebalance. For 1119 01:14:01,492 --> 01:14:08,032 trend followers TPA is the best chance we've ever had to be 1120 01:14:08,032 --> 01:14:14,572 judged on what matters, our ability to deliver value at the portfolio 1121 01:14:14,572 --> 01:14:21,172 level. This is our moment if we are ready to own it.” IloveChatGPT. 1122 01:14:05,452 --> 01:14:21,172 It's fantastic. 1123 01:14:22,212 --> 01:14:26,340 I'm sure, in the not too distant future there's going to be 1124 01:14:26,340 --> 01:14:31,156 just a blank screen next to you as you press play on Rob Carver… 1125 01:14:33,012 --> 01:14:34,132 No, no, Mr. Rob Carver. 1126 01:14:34,292 --> 01:14:38,639 Mr. Rob Carver, ChatGPT trained LLM, complete with voice, 1127 01:14:38,639 --> 01:14:45,401 and yeah, off it will go. There'll be no need for me to keep 1128 01:14:45,401 --> 01:14:46,367 signing in. 1129 01:14:46,942 --> 01:14:49,942 The good news is that is not going to happen just now. You will 1130 01:14:49,942 --> 01:14:50,702 be back soon. 1131 01:14:50,942 --> 01:14:52,702 But yes Christmas episode. 1132 01:14:52,982 --> 01:15:29,332 And, actually, there is a Christmas episode coming up which 1133 01:15:29,332 --> 01:16:12,952 will be fantastic with all the co-hosts. I think maybe one won't 1134 01:16:12,952 --> 01:16:20,222 be there. Butanyways, before we get to even any of that, I think 1135 01:15:06,758 --> 01:15:18,238 that, hopefully, I'm hoping people will go to their terminals, 1136 01:15:18,238 --> 01:15:30,866 find their favorite podcast platform, and say a big thank you 1137 01:15:30,866 --> 01:15:45,790 to you by leaving a rating and review for this episode because you 1138 01:15:45,790 --> 01:15:59,566 put a lot of hard work in answering all those questions. So, 1139 01:15:59,566 --> 01:16:12,194 we really appreciate that, as well as reviewing the articles and 1140 01:16:12,194 --> 01:16:20,230 information about the TPA. So that's great. Andif you have any 1141 01:14:57,257 --> 01:15:04,097 questions, comments, otherwise, as usual, you can send 1142 01:15:04,097 --> 01:15:07,517 them to info@toptradersunplugged.com, and 1143 01:15:07,517 --> 01:15:15,212 especially if you have something for Andrew Beer because 1144 01:15:15,212 --> 01:15:26,327 he's coming up next week. This will be fun. This will be interesting. 1145 01:15:26,327 --> 01:15:38,297 It always is with Andrew. I think we may have a little bit of 1146 01:15:38,297 --> 01:15:50,267 back and forth on some of the new stuff. He's going to be talking 1147 01:15:50,267 --> 01:16:01,382 about some news that just came out, actually, in the last day or 1148 01:16:01,382 --> 01:16:11,642 two. So, join us again next week when Andrew's here and send 1149 01:16:11,642 --> 01:16:20,192 us some questions like you did for Rob this week. Anyways,from 1150 01:14:57,698 --> 01:15:25,994 Rob and me, thanks ever so much for listening. We look forward 1151 01:15:25,994 --> 01:15:56,648 to being back with you next week. And in the meantime, as always, 1152 01:15:56,648 --> 01:16:20,228 take care of yourself and take care of each other. 1153 01:16:20,242 --> 01:16:30,457 Thanks for listening to the Systematic Investor podcast series. 1154 01:16:30,457 --> 01:16:46,347 If you enjoy this series, go on over to iTunes and leave an honest 1155 01:16:46,347 --> 01:17:01,102 rating and review. And be sure to listen to all the other episodes 1156 01:17:01,102 --> 01:17:05,642 from Top Traders Unplugged. Ifyou have questions about systematic 1157 01:16:29,326 --> 01:16:47,494 investing, send us an email with the word ‘question’ in the subject 1158 01:16:47,494 --> 01:16:53,550 line to info@toptradersunplugged.com and 1159 01:16:53,550 --> 01:17:05,662 we'll try to get it on the show. Andremember, all the discussion 1160 01:16:22,882 --> 01:16:28,690 that we have about investment performance is about the past and 1161 01:16:28,690 --> 01:16:33,442 past performance does not guarantee or even infer anything 1162 01:16:33,442 --> 01:16:38,194 about future performance. Also, understand that there's a significant 1163 01:16:38,194 --> 01:16:43,474 risk of financial loss with all investment strategies and you 1164 01:16:43,474 --> 01:16:49,282 need to request and understand the specific risks from the investment 1165 01:16:49,282 --> 01:16:54,034 manager about their products before you make investment decisions. 1166 01:16:54,034 --> 01:17:00,370 Thanks for spending some of your valuable time with us and we'll 1167 01:17:00,370 --> 01:17:05,650 see you on the next episode of the Systematic Investor.