1 00:00:01,440 --> 00:00:12,321 Welcome to Top Traders Unplugged. In markets success doesn’t 2 00:00:12,321 --> 00:00:24,411 come from predicting what happens next, it comes from being 3 00:00:24,411 --> 00:00:31,665 prepared for what you can’t predict. In each episode we go deep 4 00:00:04,620 --> 00:00:10,450 with some of the world’s most thoughtful minds in investing, economics, 5 00:00:10,450 --> 00:00:16,280 and beyond to understand how they think, how they prepare, and 6 00:00:16,280 --> 00:00:21,580 how they decide, and the experiences that shaped how they 7 00:00:21,580 --> 00:00:26,880 see the world. No noise, no short-cuts, just real conversations 8 00:00:26,880 --> 00:00:31,650 to help you think better and invest with confidence. 9 00:00:35,660 --> 00:00:38,597 Welcome back to Top Traders Unplugged where each week we take 10 00:00:38,597 --> 00:00:41,534 the pulse of the markets from the perspective of a rules-based 11 00:00:41,534 --> 00:00:44,738 investor. My name is Alan Dunne and this week I'm sitting in 12 00:00:44,738 --> 00:00:48,209 for Niels who's away on his travels and I'm delighted to be joined 13 00:00:48,209 --> 00:00:50,345 by Nick Baltas. Nick, how are you doing? 14 00:00:51,100 --> 00:00:53,256 I am doing well, Alan. I'm doing well. How are you? 15 00:00:53,500 --> 00:00:55,924 Good. I know you've been traveling as well, so there was a 16 00:00:55,924 --> 00:00:58,348 bit of tension earlier, whether you going to make it or not, 17 00:00:58,348 --> 00:00:59,560 but you're here in good shape. 18 00:00:59,580 --> 00:01:02,836 But I'm here. It's been quite a good September, obviously back 19 00:01:02,836 --> 00:01:06,980 to school, back to work, it's been very busy - the kids going back 20 00:01:06,980 --> 00:01:10,828 to school and my son starts nursing, all that lot. So, it's been 21 00:01:10,828 --> 00:01:14,676 a busy month, I must say, but all very good, all very good. 22 00:01:14,680 --> 00:01:17,992 Good. Yeah, well, it's busy and a positive month for trend following, 23 00:01:17,992 --> 00:01:21,304 so that always helps as well. So, everybody in the industry, I'm 24 00:01:21,304 --> 00:01:25,168 sure, is in good spirits. But we do always like to kick off with 25 00:01:25,168 --> 00:01:28,480 asking our guests what has been on their radar? What has been 26 00:01:28,480 --> 00:01:30,964 on your radar in the last number of weeks? 27 00:01:31,650 --> 00:01:47,754 So, I don't know that it’s something in my radar, but certainly 28 00:01:47,754 --> 00:02:03,858 something that just caught my attention, and I don't think it caught 29 00:02:03,858 --> 00:02:17,278 just my attention, I think it caught various people's attention 30 00:02:17,278 --> 00:02:36,066 and it was part of a small article on Nifty, a couple of days 31 00:02:36,066 --> 00:02:48,144 ago. This comes from GS, from Christian Mueller-Glissmann. He's 32 00:02:48,144 --> 00:03:00,222 running our kind of multi-asset strategy on the research 33 00:03:00,222 --> 00:03:19,010 side, and he put out a chart which is very, very simple but actually 34 00:03:19,010 --> 00:03:21,694 quite striking. So, what it shows is that, sorry, it shows basically 35 00:01:38,940 --> 00:01:46,959 the percentage of S&P 500 constituents (so, 500 stocks in the 36 00:01:46,959 --> 00:01:55,707 index) that have a negative beta to the index that they belong. 37 00:01:55,707 --> 00:02:05,184 So, naturally you would think if I belong to an index, then more 38 00:02:05,184 --> 00:02:14,661 likely I have a positive beta to it, partly because I'm part of 39 00:02:14,661 --> 00:02:21,951 the index, obviously at smaller sizes you would expect that 40 00:02:21,951 --> 00:02:29,970 this will become less direct, but also because there's a significant 41 00:02:29,970 --> 00:02:37,989 beta factor that is driving assets and stocks in this particular 42 00:02:37,989 --> 00:02:44,550 case. So, using like a three-month calculation point in 43 00:02:44,550 --> 00:02:55,485 time, it's 45%. So, about half of the stocks in S&P 500 have a negative 44 00:02:55,485 --> 00:03:03,504 beta to the index which I think is just mind-blowing. Last 45 00:03:03,504 --> 00:03:11,523 time that we had this situation was in.com, but at the 46 00:03:11,523 --> 00:03:21,729 time the number was just a tad less than 20%. It's now almost 45%. 47 00:01:31,650 --> 00:01:52,385 And obviously the background here is the fact that S&P is a cap-weighted 48 00:01:52,385 --> 00:02:08,335 index. The AI dominance is effectively building its own kind 49 00:02:08,335 --> 00:02:29,070 of mini index within the index and all the rest are becoming less 50 00:02:29,070 --> 00:02:45,020 correlated to it. So, this clustering and concentration is basically 51 00:02:45,020 --> 00:03:02,565 causing this effect. But again, that in itself is quite a 52 00:03:02,565 --> 00:03:21,705 striking image once you see it. I know you've seen the pictures. 53 00:03:21,740 --> 00:03:22,460 I've seen the pictures. 54 00:03:22,460 --> 00:03:23,940 So, you can relate, right? 55 00:03:24,100 --> 00:03:27,892 Yes, I mean, it shows a dramatic… well, what looks like an 56 00:03:27,892 --> 00:03:31,684 acceleration on the chart. I mean people can check it out. It 57 00:03:31,684 --> 00:03:35,792 was in the FT, so you'll see it online or check out Goldman's 58 00:03:35,792 --> 00:03:39,584 research. I mean is energy probably is driving it to an extent 59 00:03:39,584 --> 00:03:40,848 as well, I guess? 60 00:03:41,060 --> 00:03:46,002 It could be. I mean there is not too much kind of discussion in 61 00:03:46,002 --> 00:03:50,238 the actual paper from Grishman on that part aside from the fact 62 00:03:50,238 --> 00:03:54,121 that the market is getting concentrated and there's a lot of 63 00:03:54,121 --> 00:03:58,004 discussion as to whether bonds are defensive in this current environment 64 00:03:58,004 --> 00:04:01,887 and obviously this decoupling in the equity space is creating opportunities 65 00:04:01,887 --> 00:04:05,064 for dispersion. Some people talk about going long correlation, 66 00:04:05,064 --> 00:04:07,182 and so on, and so forth. So, that's kind of the background under 67 00:03:45,127 --> 00:03:51,518 which this was kind of projected. But that, in itself, even 68 00:03:51,518 --> 00:03:58,490 out of context, is quite a chart. So, not necessarily on my 69 00:03:58,490 --> 00:04:05,462 radar but let's say something that caught my attention in the last 70 00:04:05,462 --> 00:04:07,205 couple of days. 71 00:04:12,510 --> 00:06:29,270 So, noteworthy for sure? Yes. Well on my radar, I mean, probably 72 00:04:20,168 --> 00:04:33,296 similar in the sense of more noteworthy I would say… I've just 73 00:04:33,296 --> 00:04:44,236 been following the debate in markets and amongst analysts about 74 00:04:44,236 --> 00:04:55,176 the rise in yields. Obviously, everybody's been talking about the 75 00:04:55,176 --> 00:05:08,304 bond market, the bond markets around the world for the last number 76 00:05:08,304 --> 00:05:21,432 of weeks, but it just really strikes me it’s quite heated debate, 77 00:05:21,432 --> 00:05:34,560 at times, between analysts as to what's driving this rise in yields, 78 00:05:34,560 --> 00:05:47,688 as to whether it's… Is it economic growth, is it fiscal concerns, 79 00:05:47,688 --> 00:06:01,910 is it that people expect rates to be higher on average over time? 80 00:06:01,910 --> 00:06:13,944 Is it the competition for capital and is that being driven 81 00:06:13,944 --> 00:06:28,166 more by budget deficits or the AI spend? Is it the debasement credibility 82 00:06:28,166 --> 00:06:29,260 issue? So, all of these things, and everybody has their own 83 00:04:23,360 --> 00:04:37,465 view. I mean it is striking from my perspective. I mean, I do 84 00:04:37,465 --> 00:04:50,485 think the fiscal story is there in the background but at the 85 00:04:50,485 --> 00:05:04,590 same time what's really struck me, in the last couple of days, just 86 00:05:04,590 --> 00:05:17,610 looking at it, is the big repricing at the short end. Like 87 00:05:17,610 --> 00:05:32,800 the two-year yield has gone up 75 bips, or maybe a percent, at least 88 00:05:32,800 --> 00:05:45,820 75 bips, I would say, since the middle of August. So, there's 89 00:05:45,820 --> 00:05:59,925 been this… And also, at the short end, around the world has repriced. 90 00:05:59,925 --> 00:06:14,030 So, it's as if everybody now has taken a much more bullish view 91 00:06:14,030 --> 00:06:28,135 on global growth or else sees the conflict in Iran staying on much 92 00:06:28,135 --> 00:06:29,220 longer. But I suppose my perspective on this is this kind 93 00:04:25,910 --> 00:04:39,310 of tendency amongst analysts to view the fundamentals and then 94 00:04:39,310 --> 00:04:52,710 the market reaction as being very much linear and proportionate 95 00:04:52,710 --> 00:05:06,110 as opposed to markets are complex adaptive systems. A given 96 00:05:06,110 --> 00:05:23,530 change in the fundamentals at one time can have a big impact, other 97 00:05:23,530 --> 00:05:38,270 times can have a small impact. Marketing position matters (as we 98 00:05:38,270 --> 00:05:50,330 all know), the market dynamics, do people have barriers 99 00:05:50,330 --> 00:06:06,410 being hit, hedges, etc., all of that seems to get removed and 100 00:06:06,410 --> 00:06:22,490 people are trying to find this very linear cause and effect relationship 101 00:06:22,490 --> 00:06:29,190 which I don't think exists. But I'm curious to hear what's your 102 00:04:39,096 --> 00:05:24,672 perspective on yields? Do you lie in bed at night wondering why 103 00:05:24,672 --> 00:06:17,844 yields are rising or do you just take it as face value that they 104 00:06:17,844 --> 00:06:29,238 are trending higher? 105 00:06:29,510 --> 00:07:03,037 I mean I'm looking into my mortgage at the minimum and I'm trying 106 00:07:03,037 --> 00:07:36,564 to project what's going to happen the next time we have to do 107 00:07:36,564 --> 00:08:07,512 the mortgage. But I guess jokes aside, I mean some of them 108 00:08:07,512 --> 00:08:20,407 also have been quite substantial. To your point we've 109 00:06:33,814 --> 00:06:45,650 seen some kind of multi-year or perhaps decade records in the 110 00:06:45,650 --> 00:07:01,790 US in the UK as well. And even six months back… Is it six months? 111 00:07:01,790 --> 00:07:13,626 Maybe six months back analysts were projecting rate cuts. For those 112 00:07:13,626 --> 00:07:27,614 that have forgotten, that this was the case that was back in Feb 113 00:07:27,614 --> 00:07:40,526 or January. So, it's not that long ago that the projection was 114 00:07:40,526 --> 00:07:51,286 that we're entering into kind of this normalization stage whereby 115 00:07:51,286 --> 00:08:04,198 central banks will start kind of easing and all we're seeing now 116 00:08:04,198 --> 00:08:16,034 is actually yields reaching those kind of levels that we haven't 117 00:08:16,034 --> 00:08:20,338 seen for some time. I don't have a strong view, frankly speaking, 118 00:06:39,254 --> 00:06:51,434 and I wouldn't claim any particular expertise in that space. 119 00:06:51,434 --> 00:07:09,704 The one thing I would say, and that was like a quote I heard recently 120 00:07:09,704 --> 00:07:24,320 ,and I cannot recall exactly where this came from, but it was 121 00:07:24,320 --> 00:07:38,936 more about, it is very hard to predict not only because prediction 122 00:07:38,936 --> 00:07:51,116 is hard, but sometimes you cannot even tell today's situation. 123 00:07:51,116 --> 00:08:04,514 To your point, it is a struggle to understand, point in 124 00:08:04,514 --> 00:08:16,694 time, what is actually happening in the market, let alone 125 00:08:16,694 --> 00:08:20,348 predicting the markets. Obviously, that was not a quote that 126 00:06:39,093 --> 00:06:52,783 quants would be extremely happy about, on prediction, and so 127 00:06:52,783 --> 00:07:10,580 on, and so forth. But I think it's painting the picture of how 128 00:07:10,580 --> 00:07:22,901 complex markets are today. Whether we see geopolitical tensions 129 00:07:22,901 --> 00:07:37,960 kind of running ahead of us, whether that would have inflation 130 00:07:37,960 --> 00:07:51,650 implications, whether the inflation is more of a reaction to 131 00:07:51,650 --> 00:08:10,816 it or the cause of it. At the minimum, I think rates have been 132 00:08:10,816 --> 00:08:20,399 the strongest contributor to trend following strategies. 133 00:08:20,500 --> 00:08:21,580 Absolutely. Yeah. 134 00:08:21,580 --> 00:08:26,420 I would imagine most programs will be short and benefiting quite 135 00:08:26,420 --> 00:08:28,180 substantially month-to-date and year-to-date. 136 00:08:28,660 --> 00:09:25,276 Yeah, well that's a good segue. We do always like to give 137 00:09:25,276 --> 00:10:31,328 our performance update and as you say it is a good month for trend 138 00:10:31,328 --> 00:10:36,046 following. SocGen Trend up about 3.7% and SocGen CTA Index up 139 00:08:48,260 --> 00:09:13,740 about 3.7% as well. And year to date both SocGen Trend and SocGen 140 00:09:13,740 --> 00:09:31,380 CTA up about 15.2%, 15.25%. They're pretty much neck-and-neck. 141 00:09:31,380 --> 00:09:52,940 And as we were chatting, just before we started recording, that 142 00:09:52,940 --> 00:10:10,580 monthly number probably understates it a little bit because 143 00:10:10,580 --> 00:10:34,100 it's up to the 23rd and yesterday was also a pretty good 144 00:10:34,100 --> 00:10:36,060 day. So, [it’s been a] very good month, very good year. And as 145 00:08:47,296 --> 00:09:02,826 you say certainly this month looking, certainly, across the managers 146 00:09:02,826 --> 00:09:24,568 that we allocate to it very much, I'd say 80% of the cases concentrated 147 00:09:24,568 --> 00:09:41,651 in fixed income short positions. And at the long end as 148 00:09:41,651 --> 00:09:57,181 well, obviously, as I mentioned, there has been a very 149 00:09:57,181 --> 00:10:11,158 big repricing. So, interest rate futures, 2-year yields have 150 00:10:11,158 --> 00:10:31,347 had a big move as well as I suppose the more publicized move 151 00:10:31,347 --> 00:10:36,006 in bond markets. And I think this is often really where people 152 00:08:41,260 --> 00:08:58,060 see huge value in trend following as an element in a portfolio. 153 00:08:58,060 --> 00:09:16,260 Because if you think about it, if you've got a 60/40 portfolio and 154 00:09:16,260 --> 00:09:33,060 you've taken out some of the allocation and added it to trend 155 00:09:33,060 --> 00:09:45,660 following, that trend following exposure now is short bonds. 156 00:09:45,660 --> 00:10:01,060 So that bond short is basically taking down your core long 157 00:10:01,060 --> 00:10:17,860 position in bonds that you would have had. And by adding trend 158 00:10:17,860 --> 00:10:33,260 to your portfolio you'd probably be net short bonds in this 159 00:10:33,260 --> 00:10:36,060 move up. So, that's the kind of environment where you really see 160 00:08:49,890 --> 00:09:19,612 the value of it as well as also, in August, we had more long 161 00:09:19,612 --> 00:09:36,596 positions in commodities that were contributing. But certainly, 162 00:09:36,596 --> 00:10:02,072 this is where you're seeing that kind of dynamic element of trend 163 00:10:02,072 --> 00:10:23,302 following exposure; what it can do to a multi-asset portfolio, 164 00:10:23,302 --> 00:10:36,040 which I think is very valuable. It's been a good month 165 00:09:06,130 --> 00:10:36,058 for you guys as well. I think you were saying performance wise? 166 00:10:36,150 --> 00:11:23,175 I mean it's been a very good month, as you say. I think it's one 167 00:11:23,175 --> 00:12:03,930 of the stronger months that we've seen in a number of years probably. 168 00:10:36,150 --> 00:10:46,875 I think January was very strong this year and obviously the 169 00:10:46,875 --> 00:10:58,575 month that we went past, August. I mean the rates have significantly 170 00:10:58,575 --> 00:11:07,350 outsized the contribution from anything else. Commodities and equities, 171 00:11:07,350 --> 00:11:18,075 obviously, come second and third but at some margin, which is 172 00:11:18,075 --> 00:11:26,850 basically driven by this month's activity. So, said differently, 173 00:11:26,850 --> 00:11:36,600 the beginning of the month rates, commodity and equities were 174 00:11:36,600 --> 00:11:44,400 basically kind of equal contributors to performance. And 175 00:11:44,400 --> 00:11:55,125 now that gap has widened by some amount precisely because rates 176 00:11:55,125 --> 00:12:02,925 have doubled their contribution. So, it's been quite 177 00:12:02,925 --> 00:12:03,900 substantial. Currency is a bit more of a mixed bag, I would say, 178 00:10:47,982 --> 00:10:58,828 year to date, nothing very extreme in either direction. I think 179 00:10:58,828 --> 00:11:12,632 with the dollar moves we've seen there's a bit of a lack of trendiness, 180 00:11:12,632 --> 00:11:26,436 which was the case by the way for rates until maybe what, Q2? We 181 00:11:26,436 --> 00:11:36,296 hadn't seen, actually, rates kind of trending. It's been more 182 00:11:36,296 --> 00:11:48,128 of the last couple of months activity which, I guess, that's the 183 00:11:48,128 --> 00:11:55,030 other benefit of the diversification. Different contributors 184 00:11:55,030 --> 00:12:03,904 in the asset class at different points in time. But that's 185 00:10:39,884 --> 00:11:07,889 a very good month and a very good year so far. I think it's the 186 00:11:07,889 --> 00:11:34,027 third best, at least from some of the stats I have, I think it's 187 00:11:34,027 --> 00:12:00,165 the third best so far year in the last 15. So, 2014 obviously with 188 00:12:00,165 --> 00:12:03,899 a dollar… 189 00:12:03,930 --> 00:12:04,960 Yeah, 2022. 190 00:12:05,100 --> 00:12:08,250 2022, and then this year, so far so good. 191 00:12:09,530 --> 00:12:13,408 So far so good. There's always the last day of the month or the 192 00:12:13,408 --> 00:12:17,009 last day of the year to navigate. But as you say it has… 193 00:12:09,530 --> 00:12:17,010 194 00:12:17,170 --> 00:12:18,208 It has been a good month. 195 00:12:20,001 --> 00:12:29,689 It has been very good. And I think you make the point as well. 196 00:12:29,689 --> 00:12:39,377 I mean, yields had been in a range. If you look at the 10-year 197 00:12:39,377 --> 00:12:48,373 yield, it had been a 3.5% to 5% range since October 2022, whipsawing, 198 00:12:48,373 --> 00:12:55,293 it had been very difficult. Obviously, the market had priced 199 00:12:55,293 --> 00:13:03,597 in a recession probably three times since 2022 and now has flipped 200 00:13:03,597 --> 00:13:09,825 back towards a much more positive economic growth scenario. 201 00:13:09,825 --> 00:13:17,437 And now we're starting to see that move accelerate. So, interesting 202 00:13:17,437 --> 00:13:20,897 to see how that evolves. Did an interesting thing that struck 203 00:12:24,153 --> 00:12:35,225 me, I just looked at it a few days ago, so I'm not sure if it's 204 00:12:35,225 --> 00:12:42,837 moved much with the more recent run up. But volatility in 205 00:12:42,837 --> 00:12:49,757 the bond market has been relatively contained. It's been very 206 00:12:49,757 --> 00:12:57,369 stable. I mean the magnitude move has been sizable and, obviously, 207 00:12:57,369 --> 00:13:04,981 I think when you breach a psychological level like 5%, people 208 00:13:04,981 --> 00:13:13,977 tend to get very excited. But at the same time, from a volatility 209 00:13:13,977 --> 00:13:20,897 perspective, we're not seeing rising vol in the bond market. 210 00:13:21,251 --> 00:13:33,039 No. I mean and I'm seeing that also in the strength of the signals 211 00:13:33,039 --> 00:13:41,459 which are typically volatility scaled. I know they haven't basically 212 00:13:41,459 --> 00:13:49,037 shrunk or let's say they haven't actually moved... sorry… 213 00:13:49,037 --> 00:13:52,405 they haven't stayed flat. You could argue that the yield move you 214 00:13:25,547 --> 00:13:33,065 see, if it will come at a higher vol the net effect vol adjusted 215 00:13:33,065 --> 00:13:39,509 is like flat. We haven't seen that the case. What we're seeing 216 00:13:39,509 --> 00:13:44,879 is strengthening of those signals which attests to the point 217 00:13:44,879 --> 00:13:51,860 that the move has been greater than the volatility at which it comes 218 00:13:51,860 --> 00:13:52,397 along. 219 00:13:53,001 --> 00:13:57,465 Yes, yeah, obviously, and as people maintain those positions, 220 00:13:57,465 --> 00:14:03,417 the P&L attribution is that much greater than if you were seeing 221 00:14:03,417 --> 00:14:08,873 a very choppy and volatile run up. So that's favorable too. 222 00:14:08,901 --> 00:14:11,451 Less equities because equities have been a bit more challenging 223 00:14:11,451 --> 00:14:14,511 for the month. Right? Because you came into the month long and 224 00:14:14,511 --> 00:14:17,316 then, obviously, most of the markets have kind of retracted. Not 225 00:14:17,316 --> 00:14:20,121 by a significant amount but that's not like a positive month 226 00:14:20,121 --> 00:14:23,181 for equities, not too negative either. I think like less than a 227 00:14:23,181 --> 00:14:25,221 percentage, for example, I'm seeing some stats now. 228 00:14:25,591 --> 00:14:28,311 And commodities had been a standout in August? 229 00:14:28,311 --> 00:14:32,302 Mixed bag, a mixed bag for the month. So, you have some, obviously, 230 00:14:32,302 --> 00:14:35,679 some of the energy markets kind of performing but then some 231 00:14:35,679 --> 00:14:39,363 of the kind of more ags and softs have not actually delivered 232 00:14:39,363 --> 00:14:41,819 any performance or maybe detracted some performance, like 233 00:14:41,819 --> 00:14:44,889 flat for the month. It's actually quite striking that the 234 00:14:44,889 --> 00:14:48,573 month was like, flat, flat, flat except for ags, I should say. 235 00:14:48,861 --> 00:14:50,701 Yeah, well that's pretty much… 236 00:14:50,701 --> 00:14:53,925 That's the story by the way. That's the only story. It's a rates 237 00:14:53,925 --> 00:14:54,173 story. 238 00:14:54,181 --> 00:14:57,919 That's about it. 2022, I guess was a rates story, but it was also 239 00:14:57,919 --> 00:15:00,055 a dollar story as well. So, that accentuated. 240 00:15:00,301 --> 00:15:04,656 It was a bit of commodity story in Q1 as well with inflation 241 00:15:04,656 --> 00:15:07,336 moves and, obviously, the geopolitical tensions as well. 242 00:15:13,341 --> 00:15:19,403 I want to move on to some of the topics that we had prepared. 243 00:15:19,403 --> 00:15:25,032 And, I suppose, on the theme of that mix of volatility and trend 244 00:15:25,032 --> 00:15:29,362 and how to condition trend with volatility, there's a paper, 245 00:15:29,362 --> 00:15:34,558 a research note that came from the guys at Alpha Architect recently 246 00:15:34,558 --> 00:15:39,754 around the whole area of whether you can enhance a trend system 247 00:15:39,754 --> 00:15:44,517 or a momentum model by looking at the volatility environment. So, 248 00:15:44,517 --> 00:15:49,713 basically, the idea of trading faster in higher vol markets or trading 249 00:15:49,713 --> 00:15:55,342 slower in lower vol markets. You had a chance to go through the 250 00:15:55,342 --> 00:15:55,775 paper? 251 00:15:56,151 --> 00:16:07,617 Yeah, I did take a look and thanks for pointing it out. I mean 252 00:16:07,617 --> 00:16:14,988 the results are actually quite straightforward and the methodology, 253 00:16:14,988 --> 00:16:25,635 I have some kind of question marks around it. But, as you say, 254 00:16:25,635 --> 00:16:34,644 the punchline is should you adjust your kind of lookback window 255 00:16:34,644 --> 00:16:44,472 as a function of vol? If vol increases, should you go faster? 256 00:16:44,472 --> 00:16:55,119 If it drops, should you go slower? Can I reduce some of the 257 00:16:55,119 --> 00:17:04,128 false positives? Does that bring any value? That's kind of the 258 00:17:04,128 --> 00:17:06,585 high level question. But if you like the approach, it is actually 259 00:16:01,911 --> 00:16:09,591 very simple and there could be nuances related to the results that 260 00:16:09,591 --> 00:16:17,271 come from the design choices, and it's unclear to me whether we 261 00:16:17,271 --> 00:16:26,231 can generalize, but to kind of go a bit step by step, they're looking 262 00:16:26,231 --> 00:16:35,191 into… And by the way, the whole idea is that this was a research 263 00:16:35,191 --> 00:16:42,871 paper done in September 2017. So, they wanted to now extend with 264 00:16:42,871 --> 00:16:51,191 nine new years, if you like, in the sample, and see whether this 265 00:16:51,191 --> 00:17:00,791 is kind of VIX dependent. They use VIX as a measure of all VIX dependent 266 00:17:00,791 --> 00:17:06,551 lookback windows bringing more value vis-à-vis a static window. 267 00:17:07,001 --> 00:17:07,481 Yes. 268 00:17:07,801 --> 00:18:38,782 So, it's this extension of this kind of research nine years 269 00:18:38,782 --> 00:20:09,763 later. So, the whole idea is about allocating into nine… sorry… 270 00:20:09,763 --> 00:20:34,576 into five ETFs. So, they have (and I'm kind of checking the list) 271 00:17:36,121 --> 00:18:01,609 S&P, US equities outside of S&P, international equities, global 272 00:18:01,609 --> 00:18:32,761 IG bonds, and then short-term US treasuries, those five ETFs. And 273 00:18:32,761 --> 00:19:06,745 the way that they design this, let's call it trend following strategy, 274 00:19:06,745 --> 00:19:43,561 is that they choose either based on a dynamic window or on a 275 00:19:43,561 --> 00:20:20,377 fixed window, either the best or the two best assets out of the 276 00:20:20,377 --> 00:20:34,537 five based on their momentum. Now a few details here. This is, 277 00:17:26,358 --> 00:17:52,868 to a certain extent, more of a cross-sectional momentum strategy. 278 00:17:52,868 --> 00:18:27,331 Because if you're picking the best, it is not clear that this is 279 00:18:27,331 --> 00:19:04,445 a long or a short position, but it just happens to be the better 280 00:19:04,445 --> 00:19:36,257 of five. And obviously the same applies for the two. There is 281 00:19:36,257 --> 00:20:00,116 no shorting, so it's a long-only portfolio. You're always 282 00:20:00,116 --> 00:20:29,277 long the outperformer of the five or the two outperformers of 283 00:20:29,277 --> 00:20:34,579 the five. So, we can just draw the parallels or maybe the differences 284 00:18:18,883 --> 00:19:29,965 to how trend followers would look at it from a directionality, 285 00:19:29,965 --> 00:20:21,661 not just the relative performance perspective. That's point 286 00:20:21,661 --> 00:20:34,585 number one. Point number two, the fixed window uses 10 months. 287 00:17:23,569 --> 00:17:44,593 I don't know what is the 10 months. I don't really personally 288 00:17:44,593 --> 00:18:09,121 use it, but I guess it's the 200 days, that's the closest I can 289 00:18:09,121 --> 00:18:33,649 get to. It's like a 200 kind of moving average or some sort of 290 00:18:33,649 --> 00:18:58,177 a 200 days of a signal, that's the fixed. And then they have kind 291 00:18:58,177 --> 00:19:22,705 of three windows depending on the level of VIX. So, if VIX is too 292 00:19:22,705 --> 00:19:41,977 high they go very short-term. I cannot recall exactly the numbers, 293 00:19:41,977 --> 00:20:03,001 but let's say short-term. If VIX is in an intermediate range they 294 00:20:03,001 --> 00:20:27,529 go to an intermediate window and then if VIX is low they go to 295 00:20:27,529 --> 00:20:34,537 like a longer window. Now, obviously, as soon as you use kind 296 00:17:42,993 --> 00:18:22,584 of VIX triggers, everything becomes very parameterized and quite 297 00:18:22,584 --> 00:19:06,574 sensitive to the choices you're making. As an example, the 298 00:19:06,574 --> 00:20:03,761 periods of high VIX in those nine years of extension of the research 299 00:20:03,761 --> 00:20:34,554 are just three periods, three months actually. By the way, they 300 00:17:20,725 --> 00:17:56,266 do monthly rebalancing, which is another, perhaps, more of a simplified 301 00:17:56,266 --> 00:18:31,807 assumption. I mean, in all fairness, and I guess being fair 302 00:18:31,807 --> 00:19:10,579 to the approach here, possibly the audience is a bit more retail 303 00:19:10,579 --> 00:19:49,351 oriented, so they're looking at let's say five ETFs. How can I 304 00:19:49,351 --> 00:20:28,123 allocate between the five ETFs should I use kind of momentum signals 305 00:20:28,123 --> 00:20:34,585 and VIX? So, with three data points in what they call the red 306 00:17:25,566 --> 00:17:43,331 zone, I don't know how much statistical confidence we can build 307 00:17:43,331 --> 00:18:02,711 which, by the way, is fully disclaimed in the article. And then 308 00:18:02,711 --> 00:18:23,706 the last point that I would flag in this whole kind of design 309 00:18:23,706 --> 00:18:46,316 is that either you choose the top of the five or you choose the 310 00:18:46,316 --> 00:19:08,926 first two out of the five. And if you happen to choose the two, 311 00:19:08,926 --> 00:19:23,461 you would allocate 50/50 notional. So, there's no volatility 312 00:19:23,461 --> 00:19:41,226 scaling, there's nothing as such. So, it’s a very simple design 313 00:19:41,226 --> 00:20:00,606 - top one or top two, equal weights, in dollar notional terms, 314 00:20:00,606 --> 00:20:19,986 monthly rebalancing. So, there is no follow the trend, no risk scaling, 315 00:20:19,986 --> 00:20:34,521 no optimizations, no risk allocations. There's none of that. 316 00:20:34,911 --> 00:20:40,035 So, as you say, it's very much aimed at an RIA type audience. Managing 317 00:20:40,035 --> 00:20:43,695 long-only portfolios and then assessing the merits of adding a 318 00:20:43,695 --> 00:20:44,427 momentum overlay. 319 00:20:44,681 --> 00:21:08,254 Perhaps, but there are times, for example, that you might find 320 00:21:08,254 --> 00:21:31,827 one variation outperforming the other because one is in S&P and 321 00:21:31,827 --> 00:21:59,686 the other one in Treasuries. But US$100 in S&P is much more risky 322 00:21:59,686 --> 00:22:25,402 than US$100 in treasuries. So, but that is the mindset here. We're 323 00:22:25,402 --> 00:22:44,689 talking about fixed notional. It's a fixed notional allocation. 324 00:20:44,681 --> 00:20:59,837 So, with all those caveats in place, what they find, broadly speaking, 325 00:20:59,837 --> 00:21:13,730 is no significant value above and beyond a 10-month static lookback. 326 00:21:13,730 --> 00:21:31,412 There is an argument that when you choose the best out of the five, 327 00:21:31,412 --> 00:21:44,042 you're kind of outperforming. But frankly speaking, it's a period 328 00:21:44,042 --> 00:22:02,987 between 2020 and 2021 that you happen to be a bit more kind of moderate, 329 00:22:02,987 --> 00:22:18,143 that volatility wasn't too high but it wasn't too low at the 330 00:22:18,143 --> 00:22:33,299 same time. But if you look through the details, the rest of 331 00:22:33,299 --> 00:22:43,403 the periods, there is no statistical outperformance, neither 332 00:22:43,403 --> 00:22:44,666 underperformance. Obviously, the good question is what happens 333 00:21:04,218 --> 00:21:40,501 after cost? And they have some analysis on this one. When they use 334 00:21:40,501 --> 00:22:11,202 the top two, there is no significant difference at all versus 335 00:22:11,202 --> 00:22:44,694 using the same two but coming from a kind of fixed window. So, 336 00:20:45,583 --> 00:20:56,407 all that to say that there is no contradictory information to what 337 00:20:56,407 --> 00:21:07,231 I guess was the case back in 2017. Perhaps there's an argument 338 00:21:07,231 --> 00:21:14,447 that, I know without statistical confidence, there is 339 00:21:14,447 --> 00:21:26,173 some value. But as we know, with these type of studies and those 340 00:21:26,173 --> 00:21:36,997 triggers, it is hard to make a statistical case. And triggers themselves 341 00:21:36,997 --> 00:21:46,017 by nature lend themselves to very low statistical power because 342 00:21:46,017 --> 00:21:56,841 you need the trigger to be activated. And guess what? The trigger 343 00:21:56,841 --> 00:22:08,567 is not going to be activated that many times because if it's too 344 00:22:08,567 --> 00:22:19,391 many times, then it's not even a trigger. So, the whole trigger 345 00:22:19,391 --> 00:22:30,215 rabbit hole (and I'm not personally like a big fan of triggers 346 00:22:30,215 --> 00:22:40,137 in a strategy) doesn't seem, in that case, to add significant 347 00:22:40,137 --> 00:22:44,647 value. That's the whole story. 348 00:22:44,701 --> 00:22:56,485 Did selecting one condition on the VIX, it did outperform, but as 349 00:22:56,485 --> 00:23:09,251 you say, it was very much linked to, I think it was November 350 00:22:44,701 --> 00:23:09,251 2020. 351 00:23:09,251 --> 00:23:09,751 2020, exactly. 352 00:23:09,751 --> 00:23:09,927 Which was the month when the COVID vaccine was announced. And 353 00:23:09,927 --> 00:23:10,135 so, the small cap stocks, I think they were up 18% (I checked). 354 00:23:10,135 --> 00:23:10,247 I think up 10% or 11%. Yeah. 355 00:23:10,251 --> 00:23:14,031 I think I was looking into this one just a second ago. I think 356 00:23:14,031 --> 00:23:17,271 S&P at the time was like (exactly to your point) in double 357 00:23:17,271 --> 00:23:20,781 digits. And then the more kind of static strategy had like no treasuries 358 00:23:20,781 --> 00:23:21,321 at 1%. 359 00:23:22,211 --> 00:23:27,491 Yeah. I mean as you said, it’s concentration and small number of 360 00:23:27,491 --> 00:23:33,211 instances driving some of the changes. But I mean there is, I mean 361 00:23:33,211 --> 00:23:38,931 it is certainly, I suppose, an idea that I know some managers do 362 00:23:38,931 --> 00:23:43,331 model and do integrate into their trend strategies this idea 363 00:23:43,331 --> 00:23:48,611 of trading faster when market vol is higher and slower when it's 364 00:23:48,611 --> 00:23:53,891 not or when vol is lower. Have you looked at it yourself? 365 00:23:54,651 --> 00:24:15,801 So, I mean, that's the point I wanted to make as, I guess, as the 366 00:24:15,801 --> 00:24:29,901 next remark above and beyond what this article discusses. Volatility 367 00:24:29,901 --> 00:24:46,821 for us is a very important indicator and it's utilized in other 368 00:24:46,821 --> 00:25:02,331 places. So, the way that we approach portfolio design is not 369 00:25:02,331 --> 00:25:17,841 dissimilar in that mindset to utilizing volatility to be more dynamic. 370 00:25:17,841 --> 00:25:33,351 But instead of us utilizing a different kind of window, depending 371 00:25:33,351 --> 00:25:44,631 on volatility, we scale our signals by volatility. So, we look 372 00:23:58,530 --> 00:24:11,460 into signal strength that, by nature, more volatile regimes would 373 00:24:11,460 --> 00:24:26,976 bring your strength down. And if that is the case, depending as 374 00:24:26,976 --> 00:24:42,492 to how you do volatility targeting at the top level, it could 375 00:24:42,492 --> 00:24:55,422 have implications as to how much risk is eventually deployed. 376 00:24:55,422 --> 00:25:09,645 When those moves happen, I guess in a coordinated fashion, not 377 00:25:09,645 --> 00:25:22,575 only do you have volatility spikes, but you have correlation 378 00:25:22,575 --> 00:25:35,505 spiking, which also has implications on how much dollar notional 379 00:25:35,505 --> 00:25:44,556 is required for a specific volatility target. So, I'm not in 380 00:23:58,035 --> 00:24:05,649 disagreement with utilizing volatility, it's just that the way 381 00:24:05,649 --> 00:24:14,109 or the lens through which I utilize, personally, volatility in 382 00:24:14,109 --> 00:24:21,723 our research philosophy comes through the portfolio design, comes 383 00:24:21,723 --> 00:24:31,029 through the scaling of the signal, comes through the risk budgeting 384 00:24:31,029 --> 00:24:40,335 between the assets. So, it's a bit more convoluted discussion without 385 00:24:40,335 --> 00:24:49,641 necessarily having to use a trigger, or rather a continuous function 386 00:24:49,641 --> 00:24:58,101 because you're scaling by some that is continuous your estimate 387 00:24:58,101 --> 00:25:07,407 of vol, to scale your portfolio and to assess the relative 388 00:25:07,407 --> 00:25:16,713 strength of the trend signals. Obviously, we're not using VIX because 389 00:25:16,713 --> 00:25:25,173 every asset will have a different volatility, but in the 390 00:25:25,173 --> 00:25:32,787 context of utilizing market volatility, we use parts thereof 391 00:25:32,787 --> 00:25:42,939 and the correlation structure as well. So that's how I would look 392 00:25:42,939 --> 00:25:44,631 at it. 393 00:25:44,631 --> 00:25:49,023 I mean, essentially it's kind of a regime classification. 394 00:25:51,511 --> 00:25:56,935 Yes. If I come to that point, then the whole discussion becomes, 395 00:25:56,935 --> 00:26:01,907 can we identify a regime whereby a correction is more prevalent 396 00:26:01,907 --> 00:26:07,783 and is going to be more extreme? And if we could, then we 397 00:26:07,783 --> 00:26:12,755 could time trend following. So, now we're pivoting into the camp 398 00:26:12,755 --> 00:26:17,275 that says maybe we can identify trend regimes and maybe 399 00:26:17,275 --> 00:26:23,151 in that context we can utilize the volatility of the market. I mean, 400 00:26:23,151 --> 00:26:29,479 the reality is that this is a very hard gig, and I wouldn't necessarily 401 00:26:29,479 --> 00:26:33,999 argue that I can prove statistically, at least using data, 402 00:26:33,999 --> 00:26:38,971 that this is a pursuit worth pursuing, if I may say. 403 00:26:39,001 --> 00:26:44,162 I think these are all areas where people do look at from the 404 00:26:44,162 --> 00:26:47,735 perspective of trying... I mean, obviously, you mentioned using 405 00:26:47,735 --> 00:26:52,102 volatility as an input in terms of market sizing and scaling. 406 00:26:52,102 --> 00:26:56,866 But then I suppose it's then taking vol as another impulse as 407 00:26:56,866 --> 00:27:00,439 either a regime classification indicator or saying something about 408 00:27:00,439 --> 00:27:03,218 the likelihood of maybe an equity correction. 409 00:27:05,181 --> 00:27:18,341 I think my principles, and it's something I've discussed here, 410 00:27:18,341 --> 00:27:32,817 maybe with you, maybe with Niels, is that there are periods 411 00:27:32,817 --> 00:27:48,609 that we can accept the fact that predicting a reversion is hard 412 00:27:48,609 --> 00:28:05,717 or that there are no real trends in the market. And as you 413 00:28:05,717 --> 00:28:22,825 said earlier on, we've been in the right space now, in the last 414 00:28:22,825 --> 00:28:32,037 few years, with no specific direction whatsoever. So perhaps 415 00:27:07,735 --> 00:27:25,613 the prudent thing to do is to say, well look, I just succumb to 416 00:27:25,613 --> 00:27:39,660 the temptation of identifying any trend on something that is looking 417 00:27:39,660 --> 00:27:56,261 almost like white noise and I make either an active call to get 418 00:27:56,261 --> 00:28:10,308 out of some of those markets, because I'm lacking conviction, or 419 00:28:10,308 --> 00:28:21,801 I'm looking for systematic profiles that benefit from those 420 00:28:21,801 --> 00:28:32,017 regimes that would be complementary to trend following. 421 00:27:05,181 --> 00:27:14,278 So, rather than trying to make the model much more parameterized 422 00:27:14,278 --> 00:27:25,856 just to make it work, I think, to me, it's fairer to argue that 423 00:27:25,856 --> 00:27:36,607 hey, this is supposed to be doing what it's supposed to be doing 424 00:27:36,607 --> 00:27:48,185 when trends exist. If they do not exist, I'm not going to try to 425 00:27:48,185 --> 00:27:55,628 recreate them with a more complicated structure. I'd rather 426 00:27:55,628 --> 00:28:03,898 find other components that would allow me to navigate through, 427 00:28:03,898 --> 00:28:12,168 and perhaps complementing my trend allocation is the prudent thing 428 00:28:12,168 --> 00:28:22,919 to do for long term viability of the allocation. I mean that, that's 429 00:28:22,919 --> 00:28:32,016 my mantra, if you like, in this regard, from that perspective. 430 00:28:34,381 --> 00:28:38,945 It probably is fair to say though that I mean if you look at 431 00:28:38,945 --> 00:28:41,553 the performance of short-term managers and short-term momentum, 432 00:28:41,553 --> 00:28:45,465 they do tend to be stronger when you get probably higher realized 433 00:28:45,465 --> 00:28:45,791 volatility. 434 00:28:45,821 --> 00:28:54,739 For sure, for sure, but I guess my point is that is that a 435 00:28:54,739 --> 00:29:00,472 feature somebody should introduce to a more dynamic model? 436 00:29:00,472 --> 00:29:08,116 Because I would not disagree that being faster will get you better 437 00:29:08,116 --> 00:29:15,760 around turning points. But my question then is do you time the 438 00:29:15,760 --> 00:29:24,678 turning point or do you happen to be fast and when this arises you 439 00:29:24,678 --> 00:29:28,500 happen to benefit because you're there? So, you see my point? 440 00:28:50,001 --> 00:29:00,869 Calling the reversion is very hard, but if I were to be utilizing 441 00:29:00,869 --> 00:29:11,737 a fast model, I would have gotten it on time. The question is, 442 00:29:11,737 --> 00:29:20,933 can I deploy that speed dynamically, and will it be fast 443 00:29:20,933 --> 00:29:28,457 enough to cut my exposures? That's a big question. 444 00:29:28,901 --> 00:29:29,661 I hear you, yeah. 445 00:29:29,661 --> 00:29:31,061 That's a big question. 446 00:29:31,861 --> 00:29:35,677 Okay, fair enough, interesting. It sounds like you're 447 00:29:35,677 --> 00:29:41,401 not going to be making any immediate changes on the back of 448 00:29:41,401 --> 00:29:45,217 these ideas. You're pretty stable in your assessment. 449 00:29:46,111 --> 00:30:23,309 I think, I mean, I have some views, that's a fair point. I mean, 450 00:30:23,309 --> 00:30:49,879 making changes using triggers, specifically in that world of QIS, 451 00:30:49,879 --> 00:31:19,106 is not an easy path because proving statistical value is impossible, 452 00:31:19,106 --> 00:31:45,676 just because you don't have that many observations. When this 453 00:31:45,676 --> 00:32:01,618 particular article talks about three months… I was in a conversation 454 00:29:54,796 --> 00:30:08,692 recently, somebody was discussing about a specific indicator, 455 00:30:08,692 --> 00:30:26,062 or achieving two timed specific market events, and I was 456 00:30:26,062 --> 00:30:45,169 like, okay, that's great, tell me your false positives and then 457 00:30:45,169 --> 00:31:00,802 we're going to have a conversation. Because it’s great 458 00:31:00,802 --> 00:31:23,383 if you can pick the turning points. but if you pick another 30 459 00:31:23,383 --> 00:31:47,701 turning points out of the 33 in total, then I'm not so sure that 460 00:31:47,701 --> 00:32:01,597 those that you picked right are actually valid. But jokes aside, 461 00:29:51,757 --> 00:30:12,459 I think the way that the markets have been operating recently, 462 00:30:12,459 --> 00:30:35,043 in the last couple of years (and we've discussed that quite at 463 00:30:35,043 --> 00:30:55,745 length), with very short live vol spikes that moderate very quickly 464 00:30:55,745 --> 00:31:18,329 (these are all the V shapes), that we have seen, basically, one 465 00:31:18,329 --> 00:31:42,795 per year for the last four years. I think these can cause significant 466 00:31:42,795 --> 00:31:59,733 disappointment if somebody were to follow an unconstrained trend 467 00:31:59,733 --> 00:32:01,615 follower. So, in some respect, I think they do create an invitation 468 00:29:56,913 --> 00:30:09,679 for some thought to be put into how we act and interact with 469 00:30:09,679 --> 00:30:21,463 those V shaped markets that we start seeing more often. I don't 470 00:30:21,463 --> 00:30:34,229 know if that is related to the economic regime or to the current 471 00:30:34,229 --> 00:30:44,049 presidential status in global markets, but the reality is that 472 00:30:44,049 --> 00:30:54,851 we have seen them, we have seen trend following going through 473 00:30:54,851 --> 00:31:07,617 those V shapes with a bit of a scar. Thankfully, we're now here, 474 00:31:07,617 --> 00:31:22,347 a year and a bit since the last V shape or at least the last 475 00:31:22,347 --> 00:31:34,131 more aggressive V shape. I think the March event for this year 476 00:31:34,131 --> 00:31:44,933 was not as contagious as Liberation Day was, with trend following 477 00:31:44,933 --> 00:31:56,717 having done really, really well, and I think keeps on kind of 478 00:31:56,717 --> 00:32:01,627 getting attention from serious investors. 479 00:32:03,021 --> 00:32:07,619 Very good. Well, switching gears, there is a second paper, a 480 00:32:07,619 --> 00:32:11,799 paper that you highlighted. It's very interesting and very topical 481 00:32:11,799 --> 00:32:17,651 and it's in the whole area of applying agentic AI to an asset allocation 482 00:32:17,651 --> 00:32:20,995 portfolio construction process. And I mean, certainly reading 483 00:32:20,995 --> 00:32:26,847 it at a high level, it's very much an eye-opener as to the way 484 00:32:26,847 --> 00:32:31,445 things may go looking forward in terms of institutional asset allocation. 485 00:32:31,445 --> 00:32:37,715 But the paper is from Andrew Ang. Do you want to give us, I suppose, 486 00:32:37,715 --> 00:32:43,149 the overview of the paper and then get into your thoughts on it? 487 00:32:43,271 --> 00:36:07,488 Yes. I mean, obviously, as you said, we're shifting gears in a way, 488 00:36:07,488 --> 00:39:00,287 but I don't think we're shifting gears too much from how 489 00:39:00,287 --> 00:41:37,377 institutions kind of manage their assets and how asset allocation 490 00:41:37,377 --> 00:44:14,467 is perhaps getting reshaped in the world of agentic AI. So, this 491 00:33:03,305 --> 00:35:03,509 is from Andrew Ang, I would imagine he’s quite well known. He 492 00:35:03,509 --> 00:37:13,730 used to be an academic at Columbia. Then he used to be kind 493 00:37:13,730 --> 00:39:03,917 of the head of factor investing at BlackRock. And now he 494 00:39:03,917 --> 00:40:54,104 set up Altbridge with another fund, I guess of his co-founders. 495 00:40:54,104 --> 00:42:44,291 So, that's basically one of their first papers, I would imagine. 496 00:42:44,291 --> 00:44:14,444 He basically talks about agentic architecture for asset management. 497 00:32:43,271 --> 00:33:56,318 And I mean those that haven't actually seen it, and I've read it 498 00:33:56,318 --> 00:35:03,746 the last few days with a bit more detail, it's actually quite 499 00:35:03,746 --> 00:36:05,555 fascinating. I personally kind of found myself kind of drawn into 500 00:36:05,555 --> 00:37:07,364 it just because you start seeing how the transition into agentic 501 00:37:07,364 --> 00:38:03,554 AI could have implications in many things that historically we 502 00:38:03,554 --> 00:39:05,363 used to do using people, and processes, and time, and various 503 00:39:05,363 --> 00:40:01,553 parts of that process could be now abstracted. Obviously, human 504 00:40:01,553 --> 00:41:03,362 in the loop is extremely important, but then that human in 505 00:41:03,362 --> 00:42:05,171 the loop becomes more of a supervisor, more of the fiduciary 506 00:42:05,171 --> 00:43:01,361 deploying, if you like, those recommendations and having, if you 507 00:43:01,361 --> 00:44:08,789 like, a sense of the truth versus the false positives in that 508 00:44:08,789 --> 00:44:14,408 respect. So, how the whole story plays out… I'll try to summarize 509 00:35:32,231 --> 00:39:07,271 the paper if I manage. It's more of a nice story to go through 510 00:39:07,271 --> 00:42:11,591 because above and beyond it's almost like a framework. Let me just 511 00:42:11,591 --> 00:44:14,471 start by saying, what is the overall process? The overall process 512 00:33:08,558 --> 00:34:58,135 is that, let's assume you have a big institution that has to put 513 00:34:58,135 --> 00:36:22,425 together a dynamic portfolio of multiple asset classes given a 514 00:36:22,425 --> 00:37:55,144 certain level of risk, or maybe appetite, or kind of guidelines. 515 00:37:55,144 --> 00:39:36,292 And it all starts with, I guess, a policy statement. What are 516 00:39:36,292 --> 00:41:34,298 the risks we can run at? What is the volatility? What is the expected 517 00:41:34,298 --> 00:43:07,017 return? What are the asset classes we can allow ourselves to 518 00:43:07,017 --> 00:44:14,449 invest in, and so on, and so forth? So, then what happens, and 519 00:33:14,686 --> 00:34:36,365 this is now the pipeline of the agents that they're kind of proposing 520 00:34:36,365 --> 00:35:45,478 to put forward, goes as follows, like six steps. The first 521 00:35:45,478 --> 00:37:00,874 one you have, let's say a macro agent. So, the macro agent 522 00:37:00,874 --> 00:38:09,987 will do what? They will basically look into data, the economic 523 00:38:09,987 --> 00:39:06,534 regime, macro indicators, market indicators, and all this agent 524 00:39:06,534 --> 00:40:28,213 will come about is what is the economic regime, if it’s like expansion, 525 00:40:28,213 --> 00:41:31,043 late cycle, a recession, a recovery, associated with some confidence. 526 00:41:31,043 --> 00:42:33,873 And I think that's important because it's not about identifying 527 00:42:33,873 --> 00:43:49,269 a regime, but it's also about producing an assessment of the identification 528 00:43:49,269 --> 00:44:14,401 risk of that regime. Then you have asset class agents. So, now 529 00:33:24,231 --> 00:34:35,911 what you end up having is to say, well, I have 20 asset classes 530 00:34:35,911 --> 00:35:32,231 or maybe 20 assets, I don't know, US equities, global equities 531 00:35:32,231 --> 00:36:18,311 or European equities, and then treasuries, and corporate bonds, 532 00:36:18,311 --> 00:37:29,991 and real estate, and so on, and so forth. I think they run an 533 00:37:29,991 --> 00:38:31,431 example of that model for March 2026. So, I think they have 534 00:38:31,431 --> 00:39:37,991 18 assets to allocate. But let's say you have a number of agents, 535 00:39:37,991 --> 00:40:44,551 each of whom is given a specific task to focus on a specific 536 00:40:44,551 --> 00:41:51,111 asset class. And the main purpose is to come up with an expected 537 00:41:51,111 --> 00:42:52,551 return of the asset class of a volatility estimate and some confidence 538 00:42:52,551 --> 00:43:43,751 around that, and then put forward an investment case, like 539 00:43:43,751 --> 00:44:14,471 a memo, for that asset class. Then comes a, so-called covariance 540 00:33:34,091 --> 00:35:36,059 agent, which is, I would think, less of an agent and more 541 00:35:36,059 --> 00:37:38,027 of a kind of an econometric or a statistical tool that produces 542 00:37:38,027 --> 00:38:59,339 historical covariance estimation to bring this thing together. 543 00:38:59,339 --> 00:41:11,471 So, now here we are having, for all the asset classes, some projection 544 00:41:11,471 --> 00:42:32,783 of return, capital market assumptions, volatilities and covariances. 545 00:42:32,783 --> 00:44:14,423 And you need to put this thing into a portfolio. What typically 546 00:32:57,377 --> 00:34:14,960 happens is that we have one or two portfolio allocation methodologies. 547 00:34:14,960 --> 00:36:00,755 Maybe we do some sort of a mean variance or some sort of a risk 548 00:36:00,755 --> 00:37:11,285 budgeting, and that's about it. Well, the suggestion here is 549 00:37:11,285 --> 00:38:42,974 to say, why don't you actually allow the scale that agents can give 550 00:38:42,974 --> 00:40:07,610 you to run 20 portfolios? So, there are now actually there are 551 00:40:07,610 --> 00:41:32,246 19, each of whom builds a portfolio based on the previous inputs 552 00:41:32,246 --> 00:42:56,882 using different methodologies. And if I were to go through the methodologies 553 00:42:56,882 --> 00:44:14,465 (just for kind of naming a few), there's like five categories. 554 00:32:43,271 --> 00:34:46,701 The heuristics are more like no equal weighting, cap weighting, 555 00:34:46,701 --> 00:37:14,817 inverse volatility, and so on, and so forth. Some are return oriented, 556 00:37:14,817 --> 00:39:05,904 like mean variance in Black-Litterman. Some are more risk-based; 557 00:39:05,904 --> 00:40:32,305 the risk parity, max diversification, minimum volatility. 558 00:40:32,305 --> 00:42:48,078 Some are, as they call them, non-traditional, like a CVaR optimization 559 00:42:48,078 --> 00:44:14,479 or like a TPA, or drawdown constraint. And then the last category 560 00:33:37,271 --> 00:35:36,071 of all the portfolio agents has two agents which are actually 561 00:35:36,071 --> 00:37:24,071 quite special. The first one continues doing research. So, for 562 00:37:24,071 --> 00:39:22,871 the first 19, everything is almost like formulaic. Later on there 563 00:39:22,871 --> 00:41:43,271 are some tasks that they would have to undergo. But at least as 564 00:41:43,271 --> 00:42:58,871 a principle, they build portfolios, different portfolios 565 00:42:58,871 --> 00:44:14,471 that they will make available to us. So, the last kind of group 566 00:33:30,935 --> 00:35:14,207 or the last two are special. One of them is a researcher. So, 567 00:35:14,207 --> 00:36:33,647 this researcher will just do some active research, perhaps look 568 00:36:33,647 --> 00:37:45,143 into the latest innovation in the portfolio construction space 569 00:37:45,143 --> 00:39:28,415 and come up with new ideas to design a portfolio that perhaps down 570 00:39:28,415 --> 00:41:03,743 the line can become like a permanent kind of agent. So, for 571 00:41:03,743 --> 00:42:23,183 example, in this application they have is maximum entropy. So, 572 00:42:23,183 --> 00:43:42,623 this researcher is now finding maximum entropy as another possibility 573 00:43:42,623 --> 00:44:14,399 of building a portfolio. The last one, and I found it very, very 574 00:34:01,031 --> 00:35:10,151 interesting, is called an adversarial diversifier. So, this 575 00:35:10,151 --> 00:36:53,831 is now an agent that observes the rest and effectively builds a 576 00:36:53,831 --> 00:38:28,871 principal component of the rest and tries to do opposite stuff. 577 00:38:28,871 --> 00:40:12,551 Now, why is that important? Because the whole aim is to build 578 00:40:12,551 --> 00:41:56,231 debate between them and get them to a place whereby the adversarial 579 00:41:56,231 --> 00:43:22,631 diversifier is trying to find pockets of correlations that the 580 00:43:22,631 --> 00:44:14,471 rest might have not actually identified. So, now those 21 agents, 581 00:33:11,141 --> 00:34:12,455 what they end up doing, post designing the portfolios, is that 582 00:34:12,455 --> 00:35:19,343 they assess the rest and they start voting between them. They make 583 00:35:19,343 --> 00:36:09,509 some point about, obviously, whether that is computationally very 584 00:36:09,509 --> 00:37:10,823 efficient or expensive. And they say, well, we're going to force 585 00:37:10,823 --> 00:38:34,433 each agent to only vote for a peer that is in the same group and 586 00:38:34,433 --> 00:39:46,895 one outside of the same peer group. So eventually out of 21 you 587 00:39:46,895 --> 00:40:48,209 have 42 votes. The voting happens in… sorry… these are the 588 00:40:48,209 --> 00:42:00,671 reviews. Then they all vote to each other and the voting is even 589 00:42:00,671 --> 00:43:07,559 more interesting. So, every agent has to give six votes, a five, 590 00:43:07,559 --> 00:44:14,447 a four, a three, a two, a one and a minus two. Okay, and then there's 591 00:33:03,903 --> 00:33:50,325 another phase whereby all those portfolios, because they have 592 00:33:50,325 --> 00:34:41,905 now a backtest based on today's composition as inferred by 593 00:34:41,905 --> 00:35:38,643 the specific agents, also have a composite score of Sharpe ratio, 594 00:35:38,643 --> 00:36:40,539 drawdown and so on, and so forth. So, they try to basically 595 00:36:40,539 --> 00:37:42,435 score every agent now, with the voting that comes from all the 596 00:37:42,435 --> 00:38:23,699 rest, and an historical aggregate measure of performance. 597 00:38:23,699 --> 00:39:30,753 Those two combined on a 60/40 kind of scaling basis, give you the 598 00:39:30,753 --> 00:40:27,491 composite score of all the agents. So, suddenly you have now 599 00:40:27,491 --> 00:41:24,229 all the portfolio agents being informed by the asset class agents 600 00:41:24,229 --> 00:42:20,967 who talk about the capital market assumptions and come up with 601 00:42:20,967 --> 00:43:17,705 arguments in favor or against, who have already accounted for the 602 00:43:17,705 --> 00:44:14,443 macro regime. So, you can see how one brings that next. And then 603 00:32:51,125 --> 00:33:42,176 you have kind of the strategy review, which is this point of bringing 604 00:33:42,176 --> 00:34:25,373 together all those data points. You have the CRO that is 605 00:34:25,373 --> 00:35:00,716 basically the covariance agent, and then everything goes to 606 00:35:00,716 --> 00:35:59,621 the CIO. So, CIO is a new agent here, and the CIO has to produce 607 00:35:59,621 --> 00:36:54,599 now a memo for the board that is going to condense all this information 608 00:36:54,599 --> 00:37:33,869 into a recommendation and some reasoning. The interesting bit, and 609 00:37:33,869 --> 00:38:24,920 they show into some analysis, and maybe I should pause in a second 610 00:38:24,920 --> 00:39:12,044 for you to kind of start reflecting, because I know I'm going 611 00:39:12,044 --> 00:40:03,095 to keep on speaking now, in the example they show, if they were 612 00:40:03,095 --> 00:40:54,146 to run their model in March 2026, is that the CIO is actually 613 00:40:54,146 --> 00:41:37,343 getting a very different view to the voting mechanism between the 614 00:41:37,343 --> 00:42:16,613 managers… sorry… between the agents. So, the most popular allocation 615 00:42:16,613 --> 00:43:03,737 scheme in their example is not the most popular in the selection 616 00:43:03,737 --> 00:43:50,861 of the CIO. And this is primarily driven by the macro uncertainty, 617 00:43:50,861 --> 00:44:14,423 and so on, and so forth. All that to say that eventually what 618 00:33:26,468 --> 00:34:28,178 they're bringing forward is that (I'm kind of scrolling through 619 00:34:28,178 --> 00:35:42,230 my notes because this is so many things I could actually talk 620 00:35:42,230 --> 00:37:02,453 about), all they try to bring forward is to say, with GenAI… sorry… 621 00:37:02,453 --> 00:38:16,505 with agentic AI, we cannot have scale. We can have some, as 622 00:38:16,505 --> 00:39:12,044 they call it, productive dissent, which is about debating 623 00:39:12,044 --> 00:40:13,754 and peer review. So, between them, using natural language, they 624 00:40:13,754 --> 00:41:27,806 can interact. So, the third one is a natural language audit trail. 625 00:41:27,806 --> 00:42:41,858 So, if you were to go back and check how fiduciary requirements 626 00:42:41,858 --> 00:43:49,739 are actually satisfied and deployed, there is, I guess, an audit 627 00:43:49,739 --> 00:44:14,423 trail of those decisions. Then fourth kind of composable expertise. 628 00:33:57,329 --> 00:36:25,445 You put the specialist there, the CIO, some sort of forensic accounting 629 00:36:25,445 --> 00:38:53,561 module, and so on, and so forth. And then it's called governable 630 00:38:53,561 --> 00:41:09,334 delegation. So, you have the kind of the policy statement governing 631 00:41:09,334 --> 00:43:37,450 all the decisions, how much risk you should take, and so on, 632 00:43:37,450 --> 00:44:14,479 and so forth. So, to summarize, through all that process 633 00:33:21,365 --> 00:34:21,227 of agents, their point is, human beings will do the analysis 634 00:34:21,227 --> 00:35:04,763 historically, human beings will build their portfolios, and 635 00:35:04,763 --> 00:36:20,951 so on, and so forth, but the scale at which they can operate and 636 00:36:20,951 --> 00:37:31,697 the attention they can have is limited by the fact that we operate 637 00:37:31,697 --> 00:38:31,559 in real time and in human being capacities. Their point is 638 00:38:31,559 --> 00:39:20,537 that there's an element of supervision rather than replacement 639 00:39:20,537 --> 00:40:20,399 now that should allow us to elevate the level of abstraction, 640 00:40:20,399 --> 00:41:20,261 guide the process with a policy statement that a human being 641 00:41:20,261 --> 00:42:25,565 would be writing and getting us the output from the CIO agent, 642 00:42:25,565 --> 00:43:36,311 the board memo for the actual board to then decide upon, with an 643 00:43:36,311 --> 00:44:14,405 audit trail through to be fully auditable. Anyway, I'm pausing 644 00:34:32,408 --> 00:42:25,335 here. There are so many things that we can go about, but that's 645 00:42:25,335 --> 00:44:14,472 the overall summary. 646 00:44:15,201 --> 00:44:32,529 Yeah, it's very interesting. I scan through it myself and I mean, 647 00:44:32,529 --> 00:44:54,189 as you say, there's a lot of rigor to it and there's, in the sense 648 00:44:54,189 --> 00:45:05,741 of applying multiple portfolio construction techniques that wouldn't 649 00:45:05,741 --> 00:45:18,737 be typical, I'm thinking in most investment committee processes 650 00:45:18,737 --> 00:45:38,953 people maybe would use one or two or maybe a couple more. But I 651 00:45:38,953 --> 00:45:56,281 mean you're not going through… And even like in the Allocator series, 652 00:45:56,281 --> 00:46:13,609 which I host, and speaking to allocators who go through their process, 653 00:46:13,609 --> 00:46:30,937 they might say… I spoke to one recently hwo said they produced, 654 00:46:30,937 --> 00:46:46,821 I think it was three different portfolios, but here they're producing, 655 00:46:46,821 --> 00:46:56,929 I don't know, 15 or 20, effectively. I suppose what I'm curious 656 00:44:25,841 --> 00:44:51,377 about is what elements of the process is the outcome most sensitive 657 00:44:51,377 --> 00:45:14,785 to from the perspective of…? Obviously, in this example, as you 658 00:45:14,785 --> 00:45:42,449 say, they ran it in March of 2026 and from a macro perspective, 659 00:45:42,449 --> 00:46:03,729 they concluded, it was late cycle with some stagflationary risks 660 00:46:03,729 --> 00:46:27,137 which I guess reflected, obviously, what was going on in energy 661 00:46:27,137 --> 00:46:54,801 markets at that time. So that made sense. I mean, so that's one 662 00:46:54,801 --> 00:46:56,929 thing. And then obviously you have the behavior of the CIO agent. 663 00:44:36,629 --> 00:44:56,109 As you say, he tilted the allocation towards portfolio construction 664 00:44:56,109 --> 00:45:15,589 techniques that the agents didn't really vote for. But they 665 00:45:15,589 --> 00:45:31,173 were also relatively plain vanilla techniques, from memory, 666 00:45:31,173 --> 00:45:54,549 the ones that had the highest allocation. So, I didn't get enough 667 00:45:54,549 --> 00:46:17,925 detail or I didn't delve in to understand what drove his decision. 668 00:46:17,925 --> 00:46:41,301 So, I'm kind of wondering at that. Obviously there's not a rigor 669 00:46:41,301 --> 00:46:56,885 there, but ultimately what really drives the tilting? And in 670 00:44:18,795 --> 00:44:38,562 this situation the benchmark was a 60/40 portfolio, and I think 671 00:44:38,562 --> 00:45:00,126 they came out at 44% equity. So, they came out with meaningfully, 672 00:45:00,126 --> 00:45:12,705 relatively meaningfully underweight equities. Maybe the macro 673 00:45:12,705 --> 00:45:32,472 regime was part of it, I don't know. Obviously, portfolio construction 674 00:45:32,472 --> 00:45:54,036 was a bit because they come up with capital market assumptions and 675 00:45:54,036 --> 00:46:15,600 they have a mechanism. I suppose that ties to the macro regime. 676 00:46:15,600 --> 00:46:33,570 So, obviously, they put less emphasis on historic capital market 677 00:46:33,570 --> 00:46:51,540 returns and more on, I guess, forward-looking estimates. So yeah, 678 00:46:51,540 --> 00:46:56,931 that's a bit… Any insights on that or how do you think, ultimately 679 00:45:17,401 --> 00:46:19,601 that 44%, that underweight equities, what did that stem from, 680 00:46:19,601 --> 00:46:56,921 would you say, reading the paper? 681 00:46:56,941 --> 00:47:12,918 I mean it's not very obvious how it comes about because it is 682 00:47:12,918 --> 00:47:23,979 ultimately a combination of weighting schemes of different portfolio 683 00:47:23,979 --> 00:47:37,498 mechanisms. I mean, my gut feeling goes as follows, and that's 684 00:47:37,498 --> 00:47:49,788 abstracting a bit from an already abstract context here. I 685 00:47:49,788 --> 00:48:04,536 mean, ultimately the way we do asset allocation is not just informed 686 00:48:04,536 --> 00:48:16,826 by risks, and covariances, and capital market assumptions. I think 687 00:48:16,826 --> 00:48:31,574 the other statement I would make is our ability to nowcast (if 688 00:48:31,574 --> 00:48:48,780 I were to use a term we used earlier on), to nowcast the economic 689 00:48:48,780 --> 00:49:03,528 environment in a very accurate manner, then I think we are more 690 00:49:03,528 --> 00:49:13,360 prone to build some concentration in our portfolios. 691 00:49:13,360 --> 00:49:26,879 When the economic regime is more mixed, in terms of evidence 692 00:49:26,879 --> 00:49:40,398 and data points to categorize it with some confidence, we typically 693 00:49:40,398 --> 00:49:45,314 vouch more for diversification. Perhaps it's a very 694 00:47:03,741 --> 00:47:25,841 similar comment, but in a very different context to the one I made 695 00:47:25,841 --> 00:47:42,841 earlier on about trend following, reframing a model to give 696 00:47:42,841 --> 00:48:01,541 you a better estimate of what is ultimately unobservable, the current 697 00:48:01,541 --> 00:48:20,241 regime ultimately still bears the risk of the model being wrong, 698 00:48:20,241 --> 00:48:33,841 and the allocation being incorrect, and ultimately leading 699 00:48:33,841 --> 00:48:54,241 to a loss. And I posed a question here. Should we instead 700 00:48:54,241 --> 00:49:16,341 be more prone to diversify at the times that we cannot tell how 701 00:49:16,341 --> 00:49:36,741 the macro outcome is clear? So, if it's more unclear then we 702 00:49:36,741 --> 00:49:45,241 should vouch more towards diversification. So, kind of going 703 00:47:03,293 --> 00:47:23,937 back now to the paper and your question, my gut feeling is that 704 00:47:23,937 --> 00:47:39,817 the mix of kind of contradicting macro signals at the 705 00:47:39,817 --> 00:48:02,049 time, and the fact that it was like a stagflation and a late cycle 706 00:48:02,049 --> 00:48:17,929 type of a period, perhaps suggests allocation that is less 707 00:48:17,929 --> 00:48:33,809 driven by capital market assumptions. So, I think the allocation 708 00:48:33,809 --> 00:48:51,277 that the CIO goes for is less informed by return-oriented allocations 709 00:48:51,277 --> 00:49:02,393 (so like cap-weighted, volatility targeted, equal weight, 710 00:49:02,393 --> 00:49:18,273 inverse variance), these are some of the higher ranked portfolios 711 00:49:18,273 --> 00:49:37,329 that the CIO chose which do not contain any capital market assumptions. 712 00:49:37,329 --> 00:49:45,269 These are no risk-based allocations. So, in my mind it's 713 00:47:11,211 --> 00:47:39,751 more like I'm not trusting my capital market assumptions because 714 00:47:39,751 --> 00:48:16,853 the regime is not very clear and because it looks a bit fluffy. 715 00:48:16,853 --> 00:48:42,539 I should then rather tilt my allocation towards risk-oriented 716 00:48:42,539 --> 00:49:08,225 and more heuristic allocation frameworks rather than trusting the 717 00:49:08,225 --> 00:49:45,327 data as much. That's my kind of gut feeling or maybe my interpretation. 718 00:49:45,881 --> 00:49:51,718 Yeah, yeah, that makes sense. And they end up with a more conservative 719 00:49:51,718 --> 00:49:56,208 portfolio, 44% equities, which is interesting. I mean, another thing 720 00:49:56,208 --> 00:50:01,596 that was interesting that was commented upon in the paper is how 721 00:50:01,596 --> 00:50:04,739 it's very hard to do a backtest. 722 00:50:04,761 --> 00:50:07,281 I mean that's the most interesting for me. 723 00:50:07,281 --> 00:50:09,921 Backtest LLMs because they all currently incorporate information 724 00:50:09,921 --> 00:50:13,881 now that they wouldn't have had back running it in real time. 725 00:50:13,881 --> 00:50:17,841 I mean they do point out that you could… Some researchers have 726 00:50:17,841 --> 00:50:22,131 tried to limit the LLM models to use models that were only available 727 00:50:22,131 --> 00:50:26,421 at a point in time, but it sounds like that's not… Well, maybe 728 00:50:26,421 --> 00:50:29,061 those ones weren't even computationally strong enough at 729 00:50:29,061 --> 00:50:33,351 that time to run these types of analysis. So yeah, I mean, you 730 00:50:33,351 --> 00:50:38,301 have to kind of take it maybe at face value the merit a bit. I 731 00:50:38,301 --> 00:50:42,921 mean, as a researcher then how do you deal with that if you can't 732 00:50:42,921 --> 00:50:44,571 backtest, how do you evaluate? 733 00:50:45,812 --> 00:52:03,344 So that's an amazing question. That's why kind of reading it through 734 00:52:03,344 --> 00:53:27,337 put me into a lot of thoughts because there's a lot of forward-looking 735 00:53:27,337 --> 00:54:31,947 biases and they fully disclaim. So, the discussion is actually 736 00:54:31,947 --> 00:54:44,869 very nice. There is look-ahead bias by the way, in a variety of 737 00:51:18,680 --> 00:51:54,536 ways. The first one is the obvious, that you're using a model 738 00:51:54,536 --> 00:52:24,416 that was built today, perhaps to build estimates for yesterday. 739 00:52:24,416 --> 00:53:06,248 So, it has all the information baked in the model and even if you 740 00:53:06,248 --> 00:53:48,080 were to contain the data as of when, the model still has the ability 741 00:53:48,080 --> 00:54:20,948 to, or has the knowledge in its parameterization of the future, 742 00:54:20,948 --> 00:54:44,852 even if you contain, I guess, the input. The second part is, at 743 00:50:56,777 --> 00:51:20,900 the time that they start voting between them which one is 744 00:51:20,900 --> 00:51:51,602 the better model, and so on, and so forth. Then we have the typical 745 00:51:51,602 --> 00:52:17,918 look-ahead bias, or maybe data mining bias that we always have in 746 00:52:17,918 --> 00:52:42,041 backtesting. If I look into backtests, most likely I can choose 747 00:52:42,041 --> 00:53:08,357 the one that looks better. So, I would imagine the agents themselves, 748 00:53:08,357 --> 00:53:28,094 having 20 portfolio allocation schemes between them, perhaps they 749 00:53:28,094 --> 00:53:54,410 start kind of voting more in favor of the better performing one. 750 00:53:54,410 --> 00:54:16,340 So, it's almost inherent now in the process what historically 751 00:54:16,340 --> 00:54:40,463 has been human beings kind of assessing backtests. So, these are 752 00:54:40,463 --> 00:54:44,849 two considerations. The other consideration which I find even more 753 00:51:08,044 --> 00:51:30,276 interesting and, I guess, challenging, from your perspective, 754 00:51:30,276 --> 00:51:58,066 is LLM models are not deterministic, and even with the 755 00:51:58,066 --> 00:52:31,414 same model with the same prompt, you might end up having a 756 00:52:31,414 --> 00:52:53,646 different answer. So, reproducibility is nowhere to be 757 00:52:53,646 --> 00:53:26,994 seen with these type of models, at least as it stands, which 758 00:53:26,994 --> 00:53:52,005 in itself creates a significantly harder task for someone 759 00:53:52,005 --> 00:54:25,353 not only to build a backtest, but suggesting that this is representative 760 00:54:25,353 --> 00:54:44,806 of the methodology because it is not. Like, even with all the rest 761 00:51:00,752 --> 00:51:30,632 being solved, let's say today we start building a backtest as of 762 00:51:30,632 --> 00:52:00,512 today. So, tomorrow we're going to have a new model and we're 763 00:52:00,512 --> 00:52:37,862 going to do the same thing. So, we come here in a year from today, 764 00:52:37,862 --> 00:53:12,722 you and me, we have the same conversation with a year of a backtest. 765 00:53:12,722 --> 00:53:47,582 Had we rerun history or had we asked the model twice at the same 766 00:53:47,582 --> 00:54:14,972 time, a similar question with a slightly different prompt, we would 767 00:54:14,972 --> 00:54:39,872 have gotten two different backtests. And I think that's the 768 00:54:39,872 --> 00:54:44,852 other challenge. Setting that aside, I mean a few other points 769 00:51:13,748 --> 00:51:50,996 that they make, by the way, just looking through my notes, you 770 00:51:50,996 --> 00:52:28,244 might also have issues with regards to the fact that using the 771 00:52:28,244 --> 00:53:05,492 same LLM model, they call them monoculture risk. So, if all agents 772 00:53:05,492 --> 00:53:48,948 follow the same model from one of the big kind of GenAI kind of 773 00:53:48,948 --> 00:54:19,988 companies, you might find yourself being exposed to some correlation 774 00:54:19,988 --> 00:54:44,820 in the model. So, that's another interesting one. But overall, 775 00:50:50,454 --> 00:51:15,985 I think what I find interesting is the fact that there 776 00:51:15,985 --> 00:51:43,837 is here some suggestion that I would imagine can inform more accurately 777 00:51:43,837 --> 00:52:07,047 and more holistically asset owners. We cannot negate the fact 778 00:52:07,047 --> 00:52:30,257 that having the power of parallelizing all those tasks and 779 00:52:30,257 --> 00:52:55,788 creating better estimates of maybe the macro regime, or the capital 780 00:52:55,788 --> 00:53:12,035 market assumptions, or portfolio construction details, at 781 00:53:12,035 --> 00:53:35,245 the minimum should get us and investment boards more informed. 782 00:53:35,245 --> 00:54:03,097 So, in this regard, I think there is value. In the architecture 783 00:54:03,097 --> 00:54:35,591 of it, I think there is value. And above and beyond, I think the 784 00:54:35,591 --> 00:54:44,875 perspective is quite interesting. 785 00:54:44,882 --> 00:54:48,818 Just if you were to take this framework, and, obviously, it's applied 786 00:54:48,818 --> 00:54:51,442 from the perspective of longer-term asset allocation, multi-asset 787 00:54:51,442 --> 00:54:55,378 portfolio construction, but at the end of the day it's still building 788 00:54:55,378 --> 00:54:59,314 portfolios. If you were to apply it to building a trend portfolio, 789 00:54:59,314 --> 00:55:02,594 is there anything that you learned from the paper? Anything 790 00:55:02,594 --> 00:55:05,874 that you thought, okay, that's something that would be implementable 791 00:55:05,874 --> 00:55:07,186 in the near term? 792 00:55:07,642 --> 00:55:31,237 Yes. So, I think some parts are actually quite interesting because 793 00:55:31,237 --> 00:56:01,267 let's go back to this kind of food chain or workflow. So, you start 794 00:56:01,267 --> 00:56:27,007 with a… let's assume you're building a trend model and you try 795 00:56:27,007 --> 00:56:57,037 to fit it into those steps. So, you have a macro agent that says, 796 00:56:57,037 --> 00:57:20,632 okay, what is economic regime? What does the trend followers say? 797 00:57:20,632 --> 00:57:48,517 Well, I don't care about the regime, but if the price goes up, 798 00:57:48,517 --> 00:57:57,097 I'm going to buy. And obviously, the fact that I observe 799 00:55:23,896 --> 00:55:56,404 a rise in my price is nothing more than an indicator of the regime. 800 00:55:56,404 --> 00:56:24,268 Equities you'd be long more often than not when growth is there. 801 00:56:24,268 --> 00:56:54,454 And by the way, that's what economic trend does at the end of 802 00:56:54,454 --> 00:57:24,640 the day, you're looking into a regime and then you infer that your 803 00:57:24,640 --> 00:57:52,504 regime estimate is correct. So, you buy the asset that was performing 804 00:57:52,504 --> 00:57:57,148 the regime. So, pocket number one is our signal, then asset class 805 00:55:22,772 --> 00:55:37,902 agents. So, which is the expected return and the volatility 806 00:55:37,902 --> 00:55:59,084 of each one of those markets that we have, or what is the confidence 807 00:55:59,084 --> 00:56:17,240 around that? Well, I guess when we do trend following, we say 808 00:56:17,240 --> 00:56:33,883 the expected return is our past return, maybe scaled by some 809 00:56:33,883 --> 00:56:53,552 measure of vol or whatever. So, in a way the confidence we have 810 00:56:53,552 --> 00:57:11,708 is how much volatile it is, which shrinks or expands our forecast. 811 00:57:11,708 --> 00:57:29,864 And the magnitude is coming purely by how strong that trend is. 812 00:57:29,864 --> 00:57:48,020 And these are not compatible the same way as capital market assumptions 813 00:57:48,020 --> 00:57:57,098 are. So, that's step number two. So, the signal in, I guess, 814 00:55:24,316 --> 00:56:00,443 one and two are kind of similar, but I guess the macro agent 815 00:56:00,443 --> 00:56:28,233 would look into predictors of performance, whereas the asset class 816 00:56:28,233 --> 00:56:58,802 agent is more about the relative strength. The CRO, we do 817 00:56:58,802 --> 00:57:23,813 it, it's the covariance estimation. However, you want to 818 00:57:23,813 --> 00:57:54,382 estimate your covariance with a variety of techniques to make it 819 00:57:54,382 --> 00:57:57,161 robust. The fourth is portfolio construction. This is the 820 00:55:17,258 --> 00:55:32,884 place that typically we go for one or two, and that's about it. 821 00:55:32,884 --> 00:55:46,106 So, I think here there is something that somebody could spend 822 00:55:46,106 --> 00:56:01,732 some time on and looking at. Now, how that then flows into an 823 00:56:01,732 --> 00:56:13,752 investment decision in the current framework is about all the 824 00:56:13,752 --> 00:56:25,772 scoring mechanism and which portfolio favors the rest of portfolios 825 00:56:25,772 --> 00:56:40,196 and how they change their kind of their assessment of their peers. 826 00:56:40,196 --> 00:56:55,822 It’d not clear to me how that comes into trend following. But it's 827 00:56:55,822 --> 00:57:10,246 not that those steps are too abstract to how we would typically 828 00:57:10,246 --> 00:57:22,266 operate in a research environment. That's why I found it 829 00:57:22,266 --> 00:57:34,286 quite interesting because I can see, as researchers, going through 830 00:57:34,286 --> 00:57:41,498 those steps, sometimes consciously, sometimes subconsciously, 831 00:57:41,498 --> 00:57:55,922 to eventually go from raw data, to a strategy, to a single 832 00:57:55,922 --> 00:57:57,124 line. 833 00:58:00,852 --> 00:58:04,878 Is it fair to say with the evolving emerging technology that 834 00:58:04,878 --> 00:58:09,270 if you're trying to satisfy or build a portfolio for multiple objectives, 835 00:58:09,270 --> 00:58:13,662 that that's maybe easier now? Mean, just thinking, in this example, 836 00:58:13,662 --> 00:58:18,054 obviously, they had an IPS and I think it was relatively simple 837 00:58:18,054 --> 00:58:21,714 objective of whatever it was Sharpe maximization or a certain 838 00:58:21,714 --> 00:58:25,374 return for certain vol. Obviously, if you're a trend follower, 839 00:58:25,374 --> 00:58:29,034 you're trying to generate absolute return, but you want to 840 00:58:29,034 --> 00:58:33,060 tilt the portfolio to first crisis alpha or performance in a 841 00:58:33,060 --> 00:58:35,256 downturn. You want to minimize drawdowns. 842 00:58:35,322 --> 00:58:38,634 Yes. All this can come in… at times conflicting. 843 00:58:39,122 --> 00:58:43,622 Multiple objectives, at times conflicting or at times, yeah, I 844 00:58:43,622 --> 00:58:49,022 mean, hard to define. You may have more objectives than are just 845 00:58:49,022 --> 00:58:53,072 defined by mean variance optimization. That's the point, I 846 00:58:53,072 --> 00:58:58,472 guess. Does that become easier now in this kind of framework, do 847 00:58:58,472 --> 00:59:00,272 you think, or not? 848 00:59:01,741 --> 00:59:24,757 I mean, I think it becomes more accessible, let's put it that 849 00:59:24,757 --> 00:59:49,691 way. You would still have to, I would imagine, to utilize the expertise 850 00:59:49,691 --> 01:00:10,789 of running those portfolios to present something that is both sensible 851 01:00:10,789 --> 01:00:29,969 as well as tractable. Because ultimately those decisions are made 852 01:00:29,969 --> 01:00:47,231 by human beings and investors, institutional investors, that hold 853 01:00:47,231 --> 01:01:06,411 the responsibility of running those portfolios. But I think it 854 01:01:06,411 --> 01:01:23,673 will make it more accessible, just to be clear. The first paragraph 855 00:59:07,273 --> 00:59:31,245 of the paper is actually quite truthful. It says, and I'm kind of 856 00:59:31,245 --> 00:59:53,373 reading through, as you were kind of speaking, it strikes me that 857 00:59:53,373 --> 01:00:13,657 basically what you're saying is kind of in this paragraph. It 858 01:00:13,657 --> 01:00:30,253 says, “the most binding constraint in asset management is 859 01:00:30,253 --> 01:00:50,537 not data availability or model sophistication.” As in, we can get 860 01:00:50,537 --> 01:01:12,665 the data and we can be sophisticated enough, but it is the 861 01:01:12,665 --> 01:01:23,729 finite bandwidth of human decision makers. And then they go 862 00:59:10,217 --> 00:59:37,764 about talking about a CIO that perhaps can oversee 10 to 15 departments 863 00:59:37,764 --> 01:00:03,192 that are organized in lines of asset classes, and then a fundamental 864 01:00:03,192 --> 01:00:28,620 analyst can probably cover 10 to 20 stocks. And a committee that 865 01:00:28,620 --> 01:00:49,810 is kind of coming together monthly or quarterly, sometimes for 866 01:00:49,810 --> 01:01:19,476 a few hours, to just say, okay, let's go and see you in the 867 01:01:19,476 --> 01:01:23,714 next committee. It reminded me actually of a client that said, around 868 00:59:11,811 --> 00:59:22,888 the inflation regimes in 2022, “why trend following is useful: Above 869 00:59:22,888 --> 00:59:33,965 all empirical data points it is going to take opportunistic shorts 870 00:59:33,965 --> 00:59:45,042 or longs, depending on where you are, without expecting the next 871 00:59:45,042 --> 00:59:55,112 investment board to say, okay, let's change the allocation.” So 872 00:59:55,112 --> 01:00:08,203 it was purely an attempt to shrink the time that it takes between 873 01:00:08,203 --> 01:00:19,280 a decision and deployment, which is a limitation in the way 874 01:00:19,280 --> 01:00:29,350 that institutions operate. Just because time is finite and it 875 01:00:29,350 --> 01:00:37,406 kind of ticks deterministically, but it takes time 876 01:00:37,406 --> 01:00:50,497 in this way. So, one of their arguments was we like trend following 877 01:00:50,497 --> 01:00:59,560 because it can be opportunistic and deploy the exposures 878 01:00:59,560 --> 01:01:09,630 faster than our ability to bring together the investment board 879 01:01:09,630 --> 01:01:22,721 and do the analysis. Because by the time we do that, it's already 880 01:01:22,721 --> 01:01:23,728 gone. 881 01:01:24,372 --> 01:01:29,148 Yeah, interesting. I mean, one final one. I mean, obviously we have, 882 01:01:29,148 --> 01:01:33,526 in our world, programs built by machine learning. I mean, this 883 01:01:33,526 --> 01:01:37,108 is framed slightly differently with distinct agents operating in 884 01:01:37,108 --> 01:01:41,486 very well-defined capacities. I mean, how soon do you think we 885 01:01:41,486 --> 01:01:45,864 will see products in the market that are clearly labeled as 886 01:01:45,864 --> 01:01:47,456 run by AI agents? 887 01:01:50,072 --> 01:01:53,372 Hard to foresee. I think the asset management industry can utilize, 888 01:01:53,372 --> 01:01:56,372 I would imagine, those technologies. I mean, we all can 889 01:01:56,372 --> 01:01:59,072 utilize the technologies, but on our side, obviously designing 890 01:01:59,072 --> 01:01:59,672 index products. 891 01:02:00,252 --> 01:02:00,732 Yes. 892 01:02:00,972 --> 01:02:07,872 The bar is higher for the obvious reasons. I think the reason 893 01:02:07,872 --> 01:02:13,622 why this is different to just utilizing historically coding and 894 01:02:13,622 --> 01:02:18,797 backtesting platforms is that the agent is ultimately speaking 895 01:02:18,797 --> 01:02:22,247 with its peers in natural language. 896 01:02:23,372 --> 01:02:24,012 Yes, exactly. 897 01:02:24,172 --> 01:02:28,930 So, we can then go back and vet, and read, and understand, and 898 01:02:28,930 --> 01:02:32,956 check the debate. And this is all becoming much more auditable 899 01:02:32,956 --> 01:02:35,884 and tractable. But that in itself is stochastic. 900 01:02:36,632 --> 01:02:37,152 Yeah. 901 01:02:37,152 --> 01:02:43,935 It's not x plus y equals to 2, if x equals 2, what is my y? And 902 01:02:43,935 --> 01:02:48,324 I think that's the beautiful challenge in this regard. But the 903 01:02:48,324 --> 01:02:53,112 benefit of the agentic that it speaks the language rather than just 904 01:02:53,112 --> 01:02:54,309 running the math. 905 01:02:55,112 --> 01:02:59,512 Okay, interesting. Yeah, definitely a wake-up call as to how 906 01:02:59,512 --> 01:03:05,232 far things have come along in terms of putting not just the use 907 01:03:05,232 --> 01:03:10,512 of AI, but multiple agents together. And as you say, when they're 908 01:03:10,512 --> 01:03:15,352 speaking to each other and some of the additional benefits that 909 01:03:15,352 --> 01:03:21,512 come with that in terms of an audit trail, etc. So, I think, yeah, 910 01:03:21,512 --> 01:03:26,352 it's definitely… It does get you thinking as to what's coming 911 01:03:26,352 --> 01:03:26,792 next. Well, let's leave it there for today. Thanks very much, 912 01:02:59,642 --> 01:03:03,719 Nick, for bringing those research topics and always fascinating 913 01:03:03,719 --> 01:03:09,155 to get your thoughts. And Niels is away next week again, so 914 01:03:09,155 --> 01:03:14,591 I'm back again next week speaking with Cem. So if you have 915 01:03:14,591 --> 01:03:19,574 questions, please send them in. But Nick, thanks for coming and 916 01:03:19,574 --> 01:03:25,010 to everybody out there, thanks for dialing in and we'll be back 917 01:03:25,010 --> 01:03:26,822 soon with more content. 918 01:03:57,592 --> 01:05:30,566 Thanks for listening to Top Traders Unplugged. If you feel you 919 01:04:02,904 --> 01:04:16,184 learned something of value from today's episode, the best way 920 01:04:16,184 --> 01:04:33,448 to stay updated is to go on over to your favorite podcast platform 921 01:04:33,448 --> 01:04:52,040 and follow the show so that you'll be sure to get all the new 922 01:04:52,040 --> 01:05:05,320 episodes as they're released. We have some amazing guests lined 923 01:05:05,320 --> 01:05:22,584 up for you and to ensure our show continues to grow, please leave 924 01:05:22,584 --> 01:05:30,552 us an honest rating and review. It only takes a minute and 925 01:04:19,048 --> 01:05:09,112 it's the best way to show us you love the podcast. We'll see you 926 01:05:09,112 --> 01:05:30,568 next time on Top Traders Unplugged. This podcast expresses 927 01:04:03,526 --> 01:04:29,240 the views of its hosts and the guests appearing on the podcast as 928 01:04:29,240 --> 01:04:54,954 of the date of its recording, and such views are subject to change 929 01:04:54,954 --> 01:05:16,712 without notice. Top Traders Unplugged do not have any duty or 930 01:05:16,712 --> 01:05:30,558 obligation to update the information contained herein. Furthermore, 931 01:03:59,859 --> 01:04:22,529 Top Traders Unplugged make no representation to its accuracy and 932 01:04:22,529 --> 01:04:47,466 it shall not be assumed that past investment performance is an 933 01:04:47,466 --> 01:05:10,136 indication of future results. Moreover, wherever there is a potential 934 01:05:10,136 --> 01:05:30,539 for profit, there is also the possibility of loss. This content 935 01:04:07,922 --> 01:04:54,407 is made available for educational purposes only and should 936 01:04:54,407 --> 01:05:30,562 not be used for any other purpose. The information contained 937 01:04:05,338 --> 01:04:33,740 in this podcast does not constitute and should not be construed 938 01:04:33,740 --> 01:05:04,724 as investment advice or an offer to sell or a solicitation to 939 01:05:04,724 --> 01:05:30,544 buy any securities or related financial instruments in any jurisdiction. 940 01:03:57,592 --> 01:04:15,176 Certain information contained herein concerning economic trends 941 01:04:15,176 --> 01:04:40,296 and performance is based on or derived from information provided 942 01:04:40,296 --> 01:05:00,392 by independent third-party sources. Top Traders Unplugged may 943 01:05:00,392 --> 01:05:25,512 believe that the sources from which such information are obtained 944 01:05:25,512 --> 01:05:30,536 are reliable. However, Top Traders Unplugged cannot guarantee 945 01:04:06,304 --> 01:04:19,372 the accuracy of such information and has not independently 946 01:04:19,372 --> 01:04:30,988 verified the accuracy or completeness of such information 947 01:04:30,988 --> 01:04:46,960 or the assumptions on which such information is based. This podcast, 948 01:04:46,960 --> 01:05:00,028 including the information contained herein, may not be reproduced, 949 01:05:00,028 --> 01:05:17,452 copied, republished or posted in whole or in part in any form, 950 01:05:17,452 --> 01:05:30,520 without the prior written consent of Top Traders Unplugged.