"Trade Like A Chimp! Unleash Your Inner Primate" by Andreas Clenow

good afternoon everyone find your seats we'll get started in just a second I'm happy to andris canal under it he comes from us he's originally from Sweden he lives in Zurich he did his early work for Reuters doing equity excuse me doing equities and he has since become a founder of more than one hedge fund more than one hedge fund he's also a prolific author including two books stocks from the move it and so shoot did I just screw my the things you try to do from memory stocks on the move and following the trend thank you very much alright so I'm the guy hit here to speak about monkeys I'm sure they stopped the race with free eyebrows but don't worry I am getting too serious point at some part of this presentation well create like a monkey skeleckians but sorry when I look down it's way too loud sorry I just realized that obviously this topic is about portfolio modeling portfolio modeling is a little bit different from traditional technical analysis type systems where people look at one market try try to find the best entry points exit points and so on for a market that's usually what you read about in retail literature in portfolio modeling we don't really care so much about the individual stock individual portfolio component it's all about the end result how the work together therefore portfolio models are usually not that's obsessed with the exact entry the exact exit stop losses and other terminology that usually is more related to traditional type exchange trading models but volume models we need to read a couple of different components first don't worry I'll get to the monkeys we need to have a stock selection mechanism with ism stocks here obviously this is presentation series some criteria for how to select which stocks we are going to buy how we build a portfolio we allocate to our portfolio how much do we buy each simple solution of course by equal amount of each but that's usually not the most clever idea we need to rebalance because otherwise our portfolio will go out and work for while you start off the year for instance with 70% stocks and 30% bonds you wait 2 months while you no longer have 70 different stocks and sort of different forms unless everything moved exact same amount and that's not going to happen and most importantly we have the benchmarking the classic problem the in stock world well what we work with benchmarks on the hedge fund side sometimes we don't the benchmark could be the zero line but traditionally it's an index of some sort we have to compare to it maybe your your closely benchmark where you have a tried tracking error budget maybe you don't but the benchmark does matter so let's look at the benchmark first it's not too inspiring is it so we have this big bull run here and you still hear the classic views of this that stocks always go up in the long run and of course they do but the long run matter is the long run is this alone here we're ten year period with no returns at all and at one point we lost over half the money that's not a very attractive trading model so 15 year period extends five more years and get nice bull markets while we only compound five percent a year and we still have that drawdown of over half now just adjusting in the index doesn't seem to be the solution if you're unhappy to live in the right era or the right type of bull market it's fine but that's quite a gamble many mutual funds or the answer or maybe they're not so this data is from the stever reporters the S&P 500 index versus active report they measure which mutual funds managed to beat the benchmark or live up to the benchmark mutual funds of course have one single purpose to make sure beat the benchmark as it turns out any given period 3 5 10 years you have 80 85 percent of funds failing to beat the benchmark this is not the fault of the fund manager this is the fault of the model of mutual funds I'm not going to spend too much of time on that you all know the problem of mutual funds and that's why the industry is increasingly moving away from them obviously we have a problem here investing with the index even if you but embody the index as it is it's not terribly attractive if by the mutual funds well we get even worse so we've got at least an 80 85 percent probability of getting something worse than the index and it brings up the obvious question how can the team self-selected emic research and a Pyrrhic la privation supporting mais by my thesis the professor Malkiel once wrote famously in the random walk down Wall Street that a blindfolded monkey throwing darts at the board so in jobs today the list of stocks from the financial papers can outperform a group of carefully selected experts I think that statement is somewhat unfair because what I'm really accomplishing my blindfold is a monkey we are worried that he's going to aim at the right stocks but empirical observation does prove the point that monkeys know how to select stocks in 1993 I was in the beginning of my financial career in Sweden we had this prestigious competition of selecting stocks beginning of every year these are not retail people hobby peoples were competing in this these are the top investment houses in Scandinavia and very prestigious prestigious competition normally the biggest the most famous investment houses and there at top analysts win 1993 the local stock on who they entered all of the 8th into the competition all of eight walked home with a victory I incidentally that turned out very bad for all of eight because couple years later he disappears in the zoo and no one really knew where he ended up ten years later vo for the headlines that for various reasons no one could explain he was sold to some zoo in Thailand where he was not treated very well we have no idea how that happened but he probably upset their own people now unfortunately Kanto pian has denied my request actually bringing monkey here apparently this one put on New York regulation about primates throwing pointy objects that's incredible crowd excited areas so we gonna have to replace them with a net with a random number generator instead we try couple Australia's have to prove my point here because you'll think I'm kidding but maybe I am but it does have a point stretching number one here random stock picking you heard this one before this is not you I'll get to the new things here we think at the beginning of every month we liquidate the whole portfolio sell everything we take 50 random stocks a number generator selects 50 stocks put equal amount of money in each of them next month sell everything do it again of course you only pick stocks from the S&P 500 members and the important part is we of course picked from the those stocks who are a member at the time if you take these storico membership into account otherwise this would not be a fair comparison how are they still well not too bad the white line you see here is the S&P 500 total return that means you have the dividends in there that's all their turn you get this white line is of course much better than the normal SMT price index which does not include dividends and still it seems did pretty well sure we got quite a span between the best one and the worst one but none of them are even close to performing as poorly as the index clearly something is going on here this pushes a step further now we have random stocks at random weights so before we put equal amount in every stock further million dollar portfolio we just just divide the million with the 50 stars equal amount of cash in each of them but the question is what a monkey really invest like that he'll probably throw the Dodge for that too so here we have no constraints at all we could have 99.9% in one stop could be any any weight at all and this is of course resets every month again we pick new stocks at new random weights their largest step spread some of this lines they had large draw downs they had quite a lot of volatility it but they performed the did better than the index random stocks random weights works fine can we push it any further what else can we do well however the random number of stocks we pick from the S&P 500 would pick a number of stocks anywhere between 5 and 300 one of them are going to 1 and 500 well that would just be silly so we still have a portfolio 5 to 300 stocks by the monthly at completely random allocation there's land worth lands now clearly this can't work it actually does yes I did do the simulations properly I'm sure someone is wondering if I really did this mess properly and you're happy to verify it you get very similar results here you see some really wild behavior most typical sometimes you end up with portfolios with very few stocks and then anything can happen we see sudden drops when something major happen in one stock that maybe it had 90% of the other profile at the time but overall even this outperform the index over this particular period in all fairness I should point out that much of this our performance comes from this period where large cats did very poorly and that brings us to the point of no it does not yet we somewhere has dragged you first skip one or almost keep the slider these are these are the rules to come up with we have a completely random portfolio with run a number of stocks and these are the old rules you need to beat index so why are fund managers not doing this or why are not outperforming well the problem is this isn't getting it this is the main problem not the only problem this is the main problem market cap wait the S&P 500 is my benchmark here will know S&P 500 has 500 stocks that's in the name but does it really the top 10 stocks has the same weight at the bottom 300 it's about 18 percent weight for the top 10 stocks in the index and about 18 percent wage for that bottom 300 stocks so why we're pretending this is me index with 500 stops it doesn't matter the bottom 300 stocks are completely irrelevant we are just buying a handful of large caps massive companies and of course if you buy you buy Apple Apple has an approximate the market cap value of the GDP of Switzerland where I currently live then the question is can that stop double again well maybe it can but you know triples can it go up four times we start hitting some sort of natural limits here don't forget that these stocks are in there because they were one small they had enormous run-up in value and price that's how they enter the index to begin with SNP 500 is in a way a momentum index stocks are there because they had great performance in the past that's why it's terminal momentum index with a very strange weighting scheme that doesn't fit normal nothing schemes but it's an omental index now this pie chart here just shows it divides the index members in pieces of 25 stocks each the top 25% this piece next 25 stocks you compare these 25 box with the ones over here and you see that it's not a diversified index and thereby there is the problem so you could do either equal weighted the monkeys did that before or I show the red-room weighted random weighted obviously you don't want to do but it still works what would make you a bit more sense could be waiting on other textures but let's look first to this one this is the just isolating the effect of just the market cap issue see the bottom line here there is one that is the index you're used to plus and dividends this is the S&P 500 Total Return Index the top one is a total return equal weighted index it's also issued by the SMP official index is even ETFs on it this just bites the same amount of put the same amount of money in each of the stocks and yes it hasn't always outperforms not every year but outperforms over time it has always outperformed over a longer period of time and it makes perfect sense it's not as big it's not as big as performance as a random strategy room so it we only explain part of the issue but maybe we can wait on something else instead and my personal favorite which is quite common on my part of the business is to wait by volatility it's really quite simple you find some sort of measurement of volatility which one doesn't particularly matter pick a simple one pick a TR if you like it's one of the simplest estimators I wanted to put the risk parity here just to see if any of the quantity audience would pick a fight with me about the iron.i replacing the volatility parity which is actually more correct the term risk is a trick you want to use in a quant crowd so this is all till to Thursday we look at what short term volatility here amounts back or something like this I believe I did then we wait the stock allocation on the inverse volatility it means that this box is more volatile we allocate less money to it because if we allocate the same amount of money to each stock it just means that the most volatile stocks will take over the portfolio as you put a million dollars in a slow-moving utility and a million dollars in in a high volatile high volatility biotech let's portfolios going to be driven by the biotech and what I want to accomplish here is to have each stock given an equal potential Volt to the portfolio I don't want to wait my portfolio or gear the portfolio towards the volatility issues so it's picking each equal this one this simulation you looking at here it's 50 random stocks again back to that but at volatility parity this is the best one best performing random strategies in yet and based on this we know that we can with such Mythology outperform the index but obviously we're not going to go tell our clients that we're going to get a run the random number generator to pick the stocks so we have to figure out a way to end up among one of the better lines here and still have it same enough story about what we do for a living so the mission is to be the better to be the better primate we need to do a portfolio model which can take the best part of the running strategy and make it into slightly more sane strategy portfolio model is usually constructed of these pieces we need to figure out the investable universe where do we start we can start with every stock in the world we have to narrow it down somehow we need to screen for certain criteria get rid of certain stocks that we don't like ranked the stocks in the criteria that's the key point pick your key really we rank the stocks on we select the top stocks and we'll figure out the way to allocate to them and we repeat again the rebalance unless they're part of the portfolio otherwise it doesn't take long before all the nice portfolio design you need to begin with is gone with different price movement that pushes a way to sort of place so you has a classic sectors that you can pick stocks by your more this one and president this point you probably recognize things you probably if you work in the business have lots of people teaching you these kind of things at the moment separate product separate ETF for instance that they buy stocks based on these value factors dividend yield factors quality factors or ten factors there are a few other ones but these are the big ones now I'm going to take one of them completely at random has no relation to the fact that I just wrote a book about one of them we're going to take momentum I also another good reason and that is the momentum of all of these are easiest to to quantify we're just dealing with price here the other ones can be quantified but the data is not as easily accessible you have more concerns with data cleaning with making time series and so on and besides momentum you could do it directly on the control piant platform as well so a since it's your sector investing and yes of course it is factor investing has become more and more impossible in the industry lately I'm sure you get enough people teaching you this I'm not going to go in and be shoes any further but such an investing or sometimes called smart beta alternative it'll cancel on others names and make up to be unique but it's basically the same thing it's about building simple portfolio models nothing clever nothing nothing to unique usually you pick a simple way to isolate a certain sector it makes for a nice clean story a nice clean portfolio component for the allocator you have no alternative weighting scheme and learn to select each selection scheme and it's really not too difficult things but it makes perfect sense no more market cap weighting so here's the model we'll try it investable universe that we stick with S&P 500 we can do it on many indices but we have to take something here at least for this presentation and let's go to index we'll try to beat we screen them by making sure that stocks has to be current members of the index important point here if you model this is you have to take into account the so-called graveyard you have to look at what stocks remembers on the English at the time even if they are delisted merged changed otherwise it means you can forget about using Yahoo Finance and similar because it doesn't cut it you need a quality data source we have to have positive momentum we don't buy momentum stocks Terron nominators box to buy so we kick out any stocks it's just not moving up we rank these stocks based on the cleaner momentum I'm only half serious about that name but anyways I will get to what that is we select from the top the top stocks on the clean on momentum volatility parity sizing simplistic model don't worry all this source code for whatever I'm going to show you here will be available it feels already available on uncontrolled insight we rebalance monthly so this is a fairly slow trading model buying stocks that move up now the cleaner momentum what is this really well it's not so complicated it's just a variation of of ways of a way to measure momentum are you could measure momentum the simple way how many percent does something move up there in the month how many percent three more what in two months but I prefer a little bit different way complicated words but we've got a crown quant crowd here so it shouldn't be concerned annualized exponential regression slope is the basis for my handle it secure oh I'll get to exactly what that means but you take this annualized exponential regression slope you multiply it by the equipment of determination the R square and you get my little number the exponential regression I'm sure you familiar with the concept but briefly linear extra linear regression measures best fit line and how much that kills up and down the slope of this line in this case per day is we're looking at daily stock data of the problem with linear expression and in linear regression here is that we get the answer in dollars or cents you can compare that across different stocks it means something very different if it stock is trading at 200 or if it's rated at 50 so we need we need to use exponential regression that measures it measures the slope in percent then we can compare it across different stocks regardless of their base price the annualized well the problem with it the regression analytic is you get a number is very difficult to relate to it you might get a slope of zero point zero zero to five percent per day you're remembering the amount of zeros and relating to these numbers not so easy so it doesn't accomplish anything from analytical analytical point of view but we analyzed and your lies it assuming two hundred fifty trading days a year and look at what we do supply in a yearly basis now the number we get done is it answers the question how many percent would you stock move up in a full year assuming it continues the exact same path as it did in the recent past now we're not we're not believing that's going to happen but it puts a number into context that we understand how much is really moving up and down oh just to mention a small here may be the linear if you've got a linear scale on a chart then the linear regression line would look straight although if you've got a log scale on your chart well then the exponential line explanation aggression line will look straight so if you're looking at charts a lot you can probably relate easier to that explanation Quizlet of determination it fits very well together with regression we mentioned that to mention that the regression slope tells you the slope of the best fit line the R square or the equipment of determination tells you how well the third the third line really fits you can always calculate a regression line even if the data has no direction or paths or relation to each other at all then you get a very no R square if you got two points all over the place like this you get a low or square number it's a poor fit you get a number near zero if you've got a very nice fit you get a number near one which is the max maximum value now since we have a number between 1 and 0 that tells you how well the line fits well we can just multiply it there is multiply it is and then we punish those stocks that are all over the place if for instance it's very high momentum because there was a takeover offer price just jumped 100 percent in a day well it's going to push the r-square down dramatically and we multiply our high regression with a low R square and we end up with a known number at the end so we get rid of some of the some of the silly situations from this and we get the much more firm slow ascent instead so here we are with here we are with a with a variance model this model starts with us in P 500 it ranks on this momentum number of talking about it has a minimum let's just put a we put a floor to this number any number and any stock that has a value less than 30 we don't even want to consider mainly because if we allow any value there we end up with stocks with a negative number in a bear market if all the if all the stocks are in a bear market like happen in 2008 there is nothing positive to choose from y 30 well you can pick it pick at a number pick 20 it doesn't make a big deal difference experiment with it there's a source code is is available and we also make sure here as a fail-safe that stock to do to be included has to be in the top 20% of the stocks available so we look at all the S&P 500 stocks we rank them and the stock has to have a positive momentum and it has to be in top 20% I remember we do this rebalancing right so if any of these changes over time say stock is no longer part of the SMP for all of the other P 500 index in the longer part of the top 20% in the index ranked on this number or if it doesn't have a momentum value of 30 or better it goes out is rebalancing action the rebalancing as well maybe I didn't mention the properly before it not only make sure we have the right stocks as in kicking out the stocks will no longer fulfill our criterias and including stocks new stocks at doe it also resets the position sizing and that's a very important part of portfolio models that you always have to reset the position size over time even if you do an equal weighting strategy you have to do this rebalancing otherwise you end up with a very odd portfolio over time so we rank the stocks based on this number we make a ranking list this is something that's very easy to build you can build an excel even not I recommend Excel because that's not the most stable environment to do it in it's a nice environment to output the data to so you can visually inspect it and work with it but doing the maths that is not there he's not the most stable we buy from the stop by from the top until we run out of capital so there's no leverage at all in the strategy and if there are enough available stocks that fulfill our criteria well we just keep buying until we are completely out of cash we allocate again according to the inverse bola in this case this implementation we use the 20-day ATR twenty days because that covers the last month basically and next month we base the ATR on the previous month and rebalance websites again once a month we look at the results and rebalance only once a month this is a pretty slow trading strategy here are the results for that model what you're looking at here is again the same lines S&P 500 total return and the top line and show us this momentum strategy this is calculated on the control key on website by the way with this code that is already in the forum so you can replicate the exact model there what you see here is first it performs quite well it's optiforce quite nicely it has a much lower drawdown but the reason it has a much lower drawdown is this cleaning criteria that we require positive sloping stocks that also means that during some bear markets all severe bear markets we start building up cash in the strategy so if you have a concern with not being invested well vision doesn't work well it does work but not as well then again buying stocks when all stocks out there are falling like knives well that might not be the best idea regardless or type of strategy you're implementing some of the other performers here obviously comes from this but as you see we are slowly off performing in the other areas as well in the test year well okay maybe not spot any given here anything can happen so this is the contour P simulation and now I didn't include this you can look at the exact numbers you happy to do so but in this single screenshot or center it's not very easy to see what's going on but it's all out there it's on the forums and I'm happy to answer questions there with how ideas how you can modify it as well but that's one interesting question left here and as we know that this now we know this strategy can perform we know that we can outperform the index but the question is well how we're doing against the monkeys because after all if this model doesn't have to fall the monkeys then we have a problem then we just spend all this time when we can just have from a non general random number generator instead here is here the result white line again as a p500 the worst performer the blue lines you see here very shades of blue those are the the best monkeys we saw before the volatility parity selecting 50 stocks muscle rebalancing black line is off the new strategy we are many times we are leading especially we have much lower draw downs even if you eliminate the rule of putting the floor on the only analytic and essentially agree to buy negative volatility that would be the least bad volatility during bear Marcus you would still see enough performance but not as much here then of course as you notice we were pretty good here for a while but we were okay but we weren't the best couple of clever chimps I think three or four of them in this simulation what you're looking here is a fifty run of each of these random simulations I showed 50 runs I did 500 of each but frankly that looks quite silly on a charge so I included fifty selectively is random selection out of the random models and making sure that you get both the best and the worst in there so you get the full span sorry no but I can get it for you not at the moment but they're doing quite well so so I guess after all of this getting to their final points you have to wonder what the point here is and well the first point is that random modelling can actually be helpful it reveals the weakness of the index itself itself indices were created to begin with to measure the health of the stock market and as such it makes sense is not the fault of the indices they were not designed as investment strategies making investment strategies based on them well that made sense when they started doing that there was a rise of the mutual funds in 60s 70s how I guess before I was in the industry obviously it made perfect sense before that you couldn't invest in the broad markets in any other way and following the indices made sense then it does not make sense now indices are great look at the indices to see what the stock market is doing the health of the market so on the health of the economy but as a less mysterious know the random monkey strategies show us just how weak this these models are or this is into the story so creating these random simulations it actually has a point because it gives you the overview of what would have happened from purely random view then you can benchmark your strategy against it if you develop it with a nice model you have lots of rules in there you run your simulations you see that you clearly outperform the market you're very happy with your strategy well give it a try and test it against a random model do you significantly outperform the random models or not are you at least in the top bucket or the random crumbs because if you're not well cans are maybe your model bottom so bright maybe there was a problem somewhere maybe they maybe you just picked up on one of the sectors that we could already see in the random model you can see if you portfolio really adds value that way if you if you're more importantly if you if your rules add value if you add complexity if you add more rules you have to get paid for it but if you have to find a clear advantage with each rule you add if there's no clear advantage it doesn't matter how much rule how much sense the rules might make if it doesn't improve the results over run the modeling then there's a problem it is very easy to make a portfolio simulation that beats the index and that is the point if these simple simulations can do it anyone can do it in the index per se is not that difficult final point of course is that no matter how good you get you will not beat all the chimps some of them are very clever as we've seen and that's okay as long as you are one of the better one of them but if you end up in the middle of the roads in the middle of the monkey strategies well and you really didn't accomplish that much with the rules and that's pretty much what I want to tell you about I notice here as I fix presentation just before going up I do not fix the URL here so at least I won't get too many spam nails because you cannot even read what's out there and there another easy guy to find I am an accessible guy so feel free to contact me if you have any have any question of this of course and I've got two books obviously that you really should go and buy right now and I'm happy to answer any questions I think we got about three more minutes if my timer is correct a little more even okay fine all right thank you very much my question is why did you not put a short screen on you we're worried about the drawdown what you're not shorting stocks yeah they might be drivin during Norway a different topic but I believe in general shorting is they're difficult most people just should not short at all it's one of those things that look easy on the paper but frankly not in the end as you might be aware I'm also active in the ECTA business where we do short stocks or most more stock indices with the trend following models on in the future space did very well in 2000 2008 during the bear market but it's a complete misunderstanding that we did well because we shorter stocks yeah we showed the stocks and yeah we made some money on that that year but that wasn't a big-ticket the big-ticket that year was being long the non agriculture commodities being long the oil really the gold being longer bonds this was a play not in the current market obviously but shorting stocks is a problem shorting in general in trend following on momentum file strategies tend to lose money or trade even over long enough time periods it's often a good idea to do it anyway in a very diversified portfolio because it improves the skewness of the other strategy over time but complicated topic sorry they sponsor so here I'm CIA right so quintile is pitching a strategy to you what's the process you put them through and what is it you want to want to see straight up when someone is pitching a strategy to you that is correct I am on two seats we have in health strategies and we do allocate externally and I realize I just said that in public in front of a lot of people who want to sell me stuff so my own fault well well we don't really allocate to emergent emerging managers we allocate to most of the larger existing managers we want to see that things have worked properly in reality we want to see the organization we want to see their obviously their offices their the infrastructure we want to see that their proper setup to take things seriously the regulatory structure believe it or not there's a lot of companies that don't take those things seriously the things they can just sit if run to the computer and trade and that's it and they're done and it's it's a decent start but that's not how you do business we need to see that things are done properly and that they are professionals and that's not only about building the models so we look at everything I might have mentioned you might have mentioned this earlier so sorry we're going from how many different strategies you actually run so this is the momentum one but how many different ones you combine a lot we have a complicated shop we do a lot of different models and lotteries and things we're not only in systematic modeling so total investment strategy as well it's very hard to give it a normal answer so let's boil it down to just these things related to what I'm talking about probably five or six but outside of this particular field we do quite a lot of other type of strategies when you were showing some of the random portfolios around performing the SMP were you accounting for transaction cost at the monthly if you re allocating even with transaction cards from the Commission it's still outperformed LP yes let me add something extra to you here before you set a combining strategies just interesting point about that that I was asked about previously today that might make sense in this context the combining strategies that's or usually a tricky topic because combining different strategies might often make perfect sense from a investment point of view but it could actually hurt your business what I mean by that is that many allocators of course back to your question prefer clean building blocks if you make a strategy you make it a fund or portfolio that you're pitching that combines the three or four different strategies you you increase the Sharpe ratio you reduce the volatility you increase the returns by this it looks great but then you go to a large institutional chief investment officer and you pitch this and essentially you're telling him that you want to do his job for him he's going to tell you that he is his job to combine these things not yours I pitched all of them individual individual product to you that he built himself and decide how much to buy of each by one or several of them you might have a better chance from a business side okay thank you very much [Applause]

Channel: Quantopian

Views: 6,586

Rating: 4.9720278 out of 5

Keywords: finance, quantitative finance, risk, risk analysis, math, statistics, algorithms, algorithmic trading

Id: mogjvSST2H0

Length: 40min 23sec (2423 seconds)

Published: Thu Mar 09 2017