A,B,C I don't actually have any posts that analyse those precise questions. There is more info on carry in the book fr question B. It's also worth reading: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2298565 GAT
Does it make sense to include a 'sticky-coin-toss' rule or buy-and-hold rule into the bootstrap? Anything that gets weighted (significantly) above gets to stay, and anything below gets thrown away?
It's a nice idea but in practice you'd end up with hardly any trading rules and I'd personally be worried about being too undiversified. GAT
Hi GAT, Most of the way through your book. Excellent stuff, complex stuff clearly explained as if it's the most logical thing in the world. Hard to do. Well done. Some questions if I may... If you were in a seat allocating to a portfolio of CTAs - how would you judge what you were being told? I'm torn between being sold the 90phds whirring away as worth the 2/20 and then the simpler side, it's a commoditised business where getting it roughly right is enough, particularly if at .5 and 0 - which is the clearing rate now. Are the phds worth it all? Or are they there to trade bigger for EMG shareholders rather than CTA investors? What's a good 'nonsense' flag? I saw systematica drop some 12% in a week last week. Lots of mutual funds in space now, all lower costs than 2/20, although fees often buried. Would be interested in how to approach discerning between them (and traditional CTA structures). There's some stuff out there on the cost of vol adjusting in trend following. In either the blog or book or both, you mention Winton cutting the vol target in half in 08 crisis. I was around as he did that - it was for system wide panic reasons and fall of exchanges and counter parties etc. fair enough, terrifying time as you note. But you can't get past the fact that it cost his investors huge amounts. Is there anything in the argument that people bought a crisis call - with all the delta and vol expansion that implies - but he delivered a call spread, cutting positions when you needed it. The returns were great - but they 'could' have been better? A toy model of his strategy shows the risk cut - which is no secret. The scale of reduced returns though I haven't seen discussed. Not sure how to view it, in general heat of the moment makes it fine - did very similar thing. But it does make you think. Have you looked at swapping futures exposure for options depending on underlying level,of implied vol? At a theoretical level trend is like an option as positions go on akin to delta expansion. So if actual implied in say crude was 2% (ludicrous I know but for example) it would make more sense to cut all futures flat price and get the delta for way less risk with options? Floor would have to be a chunk less than normal implied/realized spread inverting, but there is a number. You mention having run gtaa strategies in the past. How do you think back on these? Given your experience at Man, if I remember they were trying to find the next ahl with a few of these. Come aug 07 and then 08, they got thumped. Bayswater, Auriel, QFS etc. I wonder how much of the capping of correlation benefits in your systems is a reflection of that (e.g. instead of long 60:40 eachin aud and cad vs -100% jpy the unclnstrained models called for 200% long aud vs -100% cad and jpy. Leading on from that - How do you think of cutting strategies if they aren't working. I get the theory, acts very differently to expectations, or hypothesis/market structure alters. But in reLity it is harder. You get married to them. Allocators get married to funds. Researchers get territorial to models. Would you talk about your experience in shutting them off? From your book, and you are clearly right, one is way better off doing the basics right, not over parameterising too much and spending time on uncorrelated diversifies rather than another trend variant. What do you think of equity market neutral as a strategy post 07? that level of crowding was incredible. Also...what other hedge fund strategies would you invest in as broad styles? Things you can't do yourself? Thanks for reading, I know it's a lot of questions. A reflection on the thought provoking nature of the book and the space as a whole. Thanks Danny
Thanks a lot. That's very kind, especially coming from someone who clearly knows their proverbial onions. (It's at this point a more pushy person would politely ask for a nice review on Amazon.com... but I'm not pushy) These are all thought provoking and difficult questions to which the true answer in most cases is "I don't know", but I'll have a stab at putting down some cursory answers. The basic CTA business ought really to be a commodity. But you do need to have people who are experienced - I think it's mostly avoiding dumb mistakes so you need to have people who've made those mistakes. I'd rather see a business splurge on building up the back office so they can diversify across instruments. Overconfidence is the main nonsense flag. If someone thinks they can do Sharpe 2.0 in the CTA business because they have some fancy machine learning algo, I'd run not walk out the door. But you get people who are experienced, not dumb, and are not overconfident who happen to be Phds. You also get people who are inexperienced, dumb, and over confident who happen to have Phds. A Phd by itself doesn't guarantee anything either way. I think here there is a basic incentive conflict underlying this - should you run your fund for the best interests of the fund investors, or the fund management company? Cutting vol in a crisis: whose interest is that in? As an economist a nice model for a systematic CTA is that they are like a commitment mechanism. The whole point of giving them your money is that they will do the hard things you can't bring yourself to do; like hanging on to large but profitable positions in a market meltdown. There is also the fact that most investors don't understand the link between vol and fees. If you cut your vol, you should cut your management fee proportionally. Yes - I have seen this. The nice thing is that the guy who is hedging your options has to do your stop loss trading for you; you just sit there and watch your delta go to zero and shrug your sholuders. A nice idea to have an integrated system that trades both options and futures; taking views on vol and direction with the cheapest instruments. Similar to how a lot of global macro traders trade. But complicated to build (mentally adds to list of research ideas...). Familiar names... I used to benchmark myself against Bayswater and I remember when they blew up in 2007. It depends on what you mean by GTAA. The 'predictor' space for GTAA overlaps too much with classic CTA (so if you're a GTAA manager working inside a CTA... you've got a problem as you're constrained to using things like low frequency macro data for which the Sharpe isn't going to be wonderful. And even Bayswater had a mandate not to be correlated with AHL). Then you've got the portfolio construction, which you described. Actually I used to run GTAA in an unconstrained way, so that didn't apply. The arguments about whether you should run a different kind of portfolio construction or another are separate from the type of forecasts: you can run momentum in a long only system, and like I said GTAA unconstrained. That comes down to whether you are running this thing as an add on / overlay and you want maximum alpha with low Beta; or your only strategy run for maximum Sharpe. Actually in my case the theory marries well to what you say. You should have a high statistical bar (p-value) to include a new model, especially if it's more complicated. But this also means you should have a high statistical bar to removing a model. In most cases for slow moving strategies you need decades of evidence to be fairly confident a model is no longer working, unless it's really shocking. I'd say the easiest strategies to shut off where were live didn't match current backtest; mainly because trading costs were massively underestimated. In that case from a statistical point of view you can be very confident you've screwed up, because the distribution of trading costs has much less noise than returns. I'd say the territorial thing works in both directions. So for example if a new PM takes over a suite of models then they'll want to throw away pretty much everything the old guy did. And everything they want to put in will be a priori wonderful. [and yes, I've been on both sides of this] You need consistent internal controls to ensure a robust and consistent statistically based approach to these in or out decisions, which overcomes these natural human instincts. Yes, I still think EMN is a good strategy, but the main flaw is not over parameterisation, but running it at a fixed target vol rather than scaling risk accordingly to opportunity. And that's true of any 'hunt for yield' strategy where returns get compressed as they get crowded (see also FX carry). So you end up leveraging up more and more at the worst possible time. Yes, in theory. But there isn't much I can't do myself in the systematic space, except access OTC markets like swaps, or do something that requires very low latency (for which I don't have the kit). Does that sound arrogant? Perhaps it's my belief that the best portfolio is a set of relatively simple implementations of various ideas, with no dumb mistakes. I know enough about most of the space that I think I could avoid most dumb mistakes (again with the exception of relatively fast trading in any asset class). And in the discretionary space, I just don't have the time or the skills to do the due diligence to work out who are the good managers (because I know that a purely statistical test of alpha would never give me results strong enough that I'd be confident just on that basis alone). You're welcome. Thanks for the interesting questions. GAT
Thanks for the reply. I appreciate it. "Yes - I have seen this. The nice thing is that the guy who is hedging your options has to do your stop loss trading for you; you just sit there and watch your delta go to zero and shrug your sholuders." There is something wonderful about outsourcing the delta hedging part if it were cheap enough, and as you say allows a view on vol as well. In general a CTA position flips in my mind from long 'vega' at inception to short gamma at max position/signal strength. Matching these with actual options makes a lot of sense. Hard to get good option data to backtest though. I also think skew would be the next logical step - there are some periods when things are so skewed 1x2s or 1x1.5s are incredibly attractive - low downside, very wide windows for terminal values with very high leverage. It does mean that you are required to pin the prices in ranges at maturity of the options though, which if CTAs are being used as a crisis hedge is tougher. Moves around the return distribution in some neat ways though, generates a synthetic carry type return, better Sharpe ratio. But again only if things are very skewed. More broadly i wonder if vol has come down so much in part as a reflection of longer opening hours and increased liquidity in electronic venues which allows more constant delta hedging. If implied is the cost of option market making, and the cost of efficiently market making has come down, makes sense competition brings down implied? "As an economist a nice model for a systematic CTA is that they are like a commitment mechanism. The whole point of giving them your money is that they will do the hard things you can't bring yourself to do; like hanging on to large but profitable positions in a market meltdown." I wonder how vol targeting at position level moves one away from this goal of hanging on to positions. It goes back to the question of 'investor' first of 'business' first. Cutting positions that are going in your favour as vol jumps dramatically alters the position function - crude in 08 a good example, vol reduction took the short down by some 75% depending on how one measures it. Better Sharpe ratio for manager, but worse 'hedge' for clients? That theoretical sold option in my call spread example goes to the manager. I also often wonder about the 'why' of trend following. I've heard a few different pitches on it - paid for cognitive biases, economies trending, payment for hedge flow and liquidity etc. All have some merit. I am quite drawn to the economies also trend/commodities are cyclical argument. Leads one to think about trend following economic numbers and then apply to market prices - eg jobless claims to trade credit spreads. But nervous of the degrees of freedom problem - there just arent that many recessions and anything that gets you out before 07 looks wonderful, your eye is automatically drawn right to it. Similar problem to GTAA in general, and EMGs desire for Bayswater as a complement to AHL. Cycle is long and factors slow moving, hard to know if 'works'. "Perhaps it's my belief that the best portfolio is a set of relatively simple implementations of various ideas, with no dumb mistakes." This I agree with 1000%. Problem is that its hard to sell that to investors who are so used to getting pitches filled with machine learning and 100 phds and a dog called Cosmos. I would rather have a 10/10 simple/robust model for running personal capital, but if running a money management business i would rather a 7/10 model and a 9/10 sales force. Sad as that is. People buying hedge funds get sold the complexity as a feature, rather than a bug IMO. But hard to change. The amount of instruments traded is often thrown out here - 300 different things! - as if these are all uncorrelated or liquid enough. I saw a good chart showing 85% of the risk of the big CTAs comes from the first 30 instruments. Seems right. Makes it easier to beat as a smaller shop as well. I saw TransTrend i think with a slide on trading hanover potatoes, as if it would make any sort of meaningful difference. Small thing also - you mention the account size issues of JGBs - do you trade the mini JGBs and mini gold? I'm still working through the book - i'll post some more questions when finished if thats OK. Thanks again for replying. DannyTrees
I'm not sure there is enough evidence for a secular fall in vol of the scale you describe. If I look at VIX back to 1990 the current level is about as low as previous lows; and the last peak was actually the highest. Something like a 10 year moving average - long enough to remove cycles - would be at exactly the same level the whole time. Cutting positions when vol goes up is just good risk management, right? I don't see how that is bad for the investor. It just means you're scaling your positions optimally, which should increase the terminal wealth of the fund. But yes, that is in isolation not in the whole portfolio context. But if you're buying something specifically for disaster insurance purposes there are products, like tail protect, with far more positive skew. Agree 100%. One of the great advantages of no longer serving clients is I no longer have to justify why models work. They worked in the past (I'm confident enough in the robustness of my backtest to say that is true). I hope they will work in the future. End of story. Obviously this doesn't apply to models which worked because of some specific wrinkle in the markets that you can point to, but they are rare and not what we're talking about. I don't; I have enough capital for the full size gold and I think the mini JGB's aren't liquid enough. GAT
Hi again - Did i see you mention a second book? Do you have a link to it, or a release date? Any teasers as to the subject/content? I'm reading Systematic Trading, 2015. I assume thats the first one?
Systematic Trading is my first book. The second book, if it was a textbook, would be called "Heuristic methods for portfolio optimisation in the presence of uncertainty and costs". But apparently that wasn't snappy enough... It's more long only investment than trading. And out later this year (September?) GAT