chat with traders episode 52 the biggest secret of the best riders in the world is that they're just like everyone else however they've worked hard to learn the markets and discover what works and what doesn't but how can you hear about these journeys in getting on the strategies and tactics they use you can do it by listening to chat with traders your host Aaron Fifield what's going on traders I hope you doing right and welcome to the podcast this way episode 52 and I had the great pleasure of speaking with dr. Ernest chair while many
traders in the quantitative rainout will already be familiar with early I'll give just a quick 30 second in charge and then we'll get into the interview sorry I guess you could say he had somewhat of and unconventional introduction to trading he started out on a research team at IBM using machine learning and artificial intelligence techniques teaching computers to understand human languages before joining a prop trading group today and he has upwards of fifteen years of points in the techniques to the demand of finance and trading working with multiple problems and hedge funds across his career
and as of right now is the managing member of qts Capital Management just the top it off and he has also written two books quantitative trading and algorithmic trading and he speaks at conferences regularly some of the topics we cover gender set-aside diversification capital allocation generating strategy ideas back testing and I whole lot more and you can get a full recap on this episode at Chavez child is dot com 452 are you can find everything they are all in one place our cause let's do this on your house Aaron Falk field this is the chapters
podcast and here is my guest from Toronto Canada any chance many as 800 I'm fine thanks for talking to me no trouble whatsoever it's nice to be speaking with you thanks for doing this already came to to dig in and get some of you insight on how he'd had such great success in markets have been looking forward to this so let's begin by talking about how it all began for you firstly you graduated with a PhD in physics where did you go from there well I went to IBM research to do some something quite different
in the field of machine learning and artificial intelligence and you know about three years as the resurgent there before I move on to finance I came up with some of the tasks that you were working on their with your eye on machine learning as well focus on teaching the computer to understand human language in all of his manifestation such as machine translation speech recognition and natural language understanding meaning that understanding how the response to human greed like nowadays iPhone Siri whatever you do that's the part of the research we did and and also how to
retrive information in response to a human language really like what Google is doing today so but bear in mind that was in 1994 at that time google did not exist and that Apple does not have the iPhone yet so we were quite early early stages of this kind of applications but the interesting thing is that the group pic of purely statistical approach we you know do not have too many many people actually that really understanding quiz takes on anything everything is based on Massa's mortals like dat ik yeah I mean I can imagine a lot
has changed since since that time so share with us how did you actually get into trading well I p.m. I decided to move to live in New York City IBM so he's a great organization has a resource headquarter a really in the middle of nowhere and I more urban lifestyle so I moved to Manhattan and did much D only kind of jobs in Manhattan at that time that suitable person with my background in finance so I joined the data mining and internal consulting group that that console for many different business units however one of the
business that we consult water trading desk in san the end at that and so we have given some opportunity to develop trading models and you know to see if they could be useful to some mobile firms traders that's really my first introduction to trading of me sort of supplying machine learning algorithms for trading research for trading strategies research ik so up until that point you hadn't really had any interest or desire to really become a trader was more just days but the jobs that were available at the time is that right that's right my interest
was in machine learning and efficient engines it's not necessary in whatever domain that can be applied you I'm more interested in the in the approach the subject matter so whether the subject matter he's human language or financial data doesn't make too much difference to me I can you that's really interesting that's actually something I was going to ask you about me to it is is that still the case well I retain my interest but I more so you know evidently after focusing on one domain which is trading about how many years that would be maybe
about more than 15 seventy years obviously I'm now you know very interested in financial market so yes my my interest has narrowed considerably 21 domain which which is financed but I'm still needs and not on application products for use other than finance sure sure following your position at Morgan Stanley I mean what was the next step from there onwards well the human mind any sort of scatter a little bit after I joined name because the main sponsor of the group was hired away by Deutsche Bank to head up there global equities trading become a very
scenic sector even I mean he was saying he was the head of the UN Security trading at morgan fuck doggy bang he becomes a global head of equities trading so you know he's a step up for him but we've often speaks concept behind out we had a loss finding clients so some of our colleagues at his side to leave and go on to Credit Suisse and stopped 25 your trading cool so that's I was asked to join them you know we went off to keep operating in credit cease to apply this technique that we we
have been residing along like a good one now I usually ask this question and I'm not sure it's quite so relevant to your background but I'm gonna ask anyone you can you can give me your thoughts let's say in the period of your first five years being involved with financial markets what was some of the greatest difficulties encountered that that really challenged you well I mean as many of the recent trading model is the greatest difficulty is that you know you can can can look very good on on the back as pieces but you know
very hard to know to perform well and you know I apply these techniques to trading obviously I have died or have a lot of experience and intuition about financial markets and so I approached it you need to see radical mechanical manner and it turns out that actually almost every model that I we didn't work live trading perfectly on you know back test basis but fail so that's a big challenge I can hear that makes sense so so pretty much what you sign is that even if you do have the you know the knowledge of machine
learning and IR and those types of things applying them to financial markets it still was very beneficial to have an understanding of how at the market operates its not just as black and white as that's you know a different data set that you know working with yes because the at that time at least the keys 4444 financial markets very poss compared to other sort of machine learning data sets you know typically we would feed millions of articles to teach you to understand human language and wheres for financial data especially training they needed and not high
frequency trading but there aren't enough later really trained a system properly and so I find was necessary to really constrained systems that can sensible even with very limited to eradicate yeah that's really interesting now any help would you describe your style of trying to anyone who is completely unfamiliar with your work so share with us maybe the frequency that your trading markets and just a little bit more insight to your quantitative approach well our focus is really the intraday trading but not high frequency trading the reason for that particular time frame is dead and I
would suggest that we know that whatever result we are seeing is not pure luck trades to convince us that we are doing something correctly not not just so you know random random luck but not high fees you joining because I'm enjoying the quiet much hired investment structure at that has had its not justify so yeah we focus on the kids enjoyed a training that holds the average and we've focused initially on four acts because that's the most liquid market and and later on we added the futures models and models so and now we actually options
as well so we are constantly looking for new models and we do not restrict ourselves in terms of what markets we we we can trade on ok so trading such a wide range of markets do you find a strategy as you develop each market Abbas the different or is it sometimes the case that a strategy for stocks may also work effectively on futures the charges are vastly different you know we we I know that there are for example believe that you can trade on juice future the same way as you trade your gold futures and
the same way that you can trade oil futures confined water or power to them so we have a separate model for every future and you might say well that sounds like overfitting to the particular user well so you know that we focus on each market has a very decent operational and season we tried so and we have been running it for a while life and and it will not work on any other market just saying the markets for every spectacle of every single instrument we have a separate model for ok and I mean how do
you deal with that when you come to market such as just the equities in stocks market as well as options remain or do you have your strategies operate on a basket of stocks were they still individual strategies for individual stocks yes stars are different because we typically run stocks market neutral strategy so tomorrow so we have won 10 to select was by and so every day and so indeed the same selection model is applicable for stocks like a running multiple strategies any given time in your opinion what are the benefits of doing this and could
you give us an idea on how many different strategies are running we run over seven or eight regional some of them are large delegation so they're not equally capitalize so some of them are experimental stage which will benefit of course is diversification portfolio are subject to many risks due to leave you alone lease portfolio of course it will be subject to market risk market neutral but somehow overlooked over waited on that's the one sector of socks and underwear and the other second you know so you will have some factories and even if and and there
some other reason at an Oslo parents and for example you have fun if you're running options you know you might be you may be negative so you might be sensitive to 240 leaves only partially goes up down or you might be some interest rate goes up so every strategy has these sort of sensitivity risk quality interest rate market research factories or not and the goal of a balanced portfolio that you want to neutralize so you know and you can't do that with just one so I do have a soft spot and he's going to be
vulnerable to particular risk but if you're running multiple strategies hopefully their sensitivity neutralize each other so and simple example long shot photo neutralize market list with a long political benefit from increasing salt but negative but some of the two sides of portfolio will have practically have no risk market neutral you will not be affected by the movement of the market index and the same can be applied to any of these respective whether it's closer to the great so far and that's the goal is to be neutral to any almost disrespect and now of course we
are not as smart we cannot we really haven't figure out exactly some real circumstances some reason that we have not neutralized by some of the obvious risk we have eliminated because of a diversification yeah that's that's excellent now let's imagine an entry level has created a trading system there were single strategy and a satisfied with the results from testing is it was for them to try this without having other strategies to diversify yes I mean you you can't just sit and wait for ready to go before we start trading solely when we started fund traded
once tried yet obviously very risky portfolio and so we had to be caught on so that was unfortunate but we have lowered our leverage a lot so until we get that was advice oh yes I mean you have just one strategy trading it but you know undergone over lever yes that's very well said and now I understand your strategies are designed to catch momentum and then some also designed to capitalize on mean reverting to you have to decide which strategy taught at best for the current market conditions by turn them on and off or is
this something that's programmed into your algorithms we have not found a way to be able to predict the next month whether momentum Shardul voting so I G it is very easy of course to do it in hindsight oh geez this last month we should have just run the momentum but how do you tell next month whether you should do so we have not found a model that can do that a lot of people claim that they have these kind of regime again all the power to them but equal so what we do is that we
as I said we run multiple models same time so that know someone who will listen some will gain and hopefully the model that Blues will lose less than the ones that game that's a delicate balance of course we don't always do it right here we do have some down once we are not able to compete every month that would be wonderful but you know we didn't get to that action yet but so far has been the case that month and diamonds much less much less than the office so they do not suffer big drawdown in
a well balanced fund portfolio and so you know it's coach at least worked way he's an extent right so with that being said there any discretionary elements to your trading style but there is some discretion in terms of the allocation to the particular reason for that is that we constantly you know some of our strategies lucky enough to have work day one and although they had been drawdown never stop working so trading it for more than four years and the backbone of our fund but there are two strategies that we started and read my full
year and become quite weak and not to lose money and we need to reduce their leverage well in that sense is somewhat systematically we do we deal with them based on their their loss but they are just created now we have a new strategy you never know how much I would say that it is important to to immediately and ok just based on back tears all discharges to al-qaida based on the back test operation I think that would be included because no matter how well it back to eight overstating biased and be ready to change
so the lives and also thirdly you don't know exactly what a marketing practices trade size so careful when you start a new strategy always starting it at a very low average and gradually increase it to the proper education and that process East discretionary because I don't see that Daisy that's no way to systematically increasing usually because whatever systematic way that you can come on and the backtest already says that this is great whole battle point that we don't believe the better since we don't completely believe the back there's necessary discretion so that part in phase
of dynamically changing sides some of them are buying some of them are getting born dad Paul says he's now ask you more about backtesting actually a little bit but right now I would like to ask you a couple questions about some of the less obvious differences between discretionary and quantitative trading so let me ask you this how does profit distribution differ between discretionary and quantitative strategies and the reason I ask is because some discretionary traders will say their profits are quite lumpy so they'll say that ninety percent of profits come from 10 percent of their
trades do you find this is also the case with Quan trading strategies or is equity curve somewhat smoother well I think that the differences between these questionnaires waiting all worth explaining the differences between being the voting and momentum strategies typically the characteristics of a media hora strategy that the Prophet hour of every trade is very similar and almost every trade will be accepted a few trades very rare trades that will lose maybe ten times the typical that's the country's whether you're running a min averted by discretion I would have the same countries doing the opposite
momentum drive you home in saudi such that a lot of trees will not be successfully infected with was a little bit of money but once in awhile there will be a rare trait that will make ten times as well maybe more than typical losses so yeah I would say that the sort of the polls out of that you described really I was a differentiated by this wedding band room now each question or not I think that the the problem with discretion now there's pros and cons to this question would think about this question trading is
that you can take you to contacts macroeconomics a lot of variables that you can take into context like a major has happened did some major decision must be made in your model where you went but human can so you know that humans and say well they obviously did they may be a major the form that looming over Greece and you don't want to trade more sensible but you would never know that's the good thing about trading that you know you can contact and particularly rare events but the problem with this question trading is that each
human brain can focus on a few things at a time and research has shown that the multitasking doesn't work you do multitasking all you get is degraded performance in every time that we do so these questions very unlikely to be able to manage mornin small handful of strategies and as a result the Africa Cup will not be terribly small and that unless you're just running very high I mean 15 you but because of the lack of diversification discretionary trader it is very unlikely that you're shopping hi well divers and and indeed will look kind of
rocket and that might be why one might see different shape of there being no that's moving training is simply a consequence of that was the lack of diversification is question I'd rather run in seven innings I don't know how they would be able to do that very well said thank you for explaining that you mentioned in there that you can take old music macro events into consideration do you take some of these news events and sentiment into consideration when developing strategies or is it purely always just buy stock price well we developing the strategy we
have not be able to take into account the news but sunny now they are more and more vendors of the new sentiment data we may be able to incorporate that going forward but even so to backtest that's not incorporate new sentiment there are some exceptional you bands that we would take into account training immortalize some rare occasion would override immortal because we don't believe that the model has to dole's events that coming up and so management one of you we will not trade because of those events such as natural disaster or major economic events so
yeah that's a great question would come in because our newspaper we feel that it is prudent to sometimes override immoral or trade because of them except ok yeah I mean that makes sense now let's talk a little bit about generating I did something you probably get off the squirt regulate that what your process of finding ideas for viable trading strategies well you know I'm sort of vacuum cleaner of ideas out there in the public and you know I have a follow-on humble tree.com and these are all accounts which regularly on a daily basis operation the
headlines on new articles on finance and trade so I would always make no properly looked interesting and I will read them when I have time and get ideas from some of them I also always books on on trading volume Valley Elementary box to box depth stands with questions or background where you consume them and and then also cause of life trading oftentimes you will absolutely dominant form you can download and that's one thing that we distinguish between age a daughter that just sit in the office and research and look at previous look at data sources
a traitor who actually watching the market real time because the information sometimes you you you're you're praying would capture information when we owe money is that you would never be able to capture he feeds just backed out so I often would get inspiration from some particularly unusual during the day that can be that can be utilized nutri more so so yes all these dreams that's a really great point I'm glad you highlighted that how do you feel about academic research on market studies in that type of thing I mean I think he kind of mentioned
it there but I mean things off kind of heard about academic research is one of the precautions that you should take is that it's not necessarily realistic to replicate those results had it been taken in a lot of market in Maine is there any precum precautions that china should be aware of if rating academic research yes obviously you no one should not assume that these research are comical and their main reason they may not turn out to be more some of them because they being over it pager others because you know they have not taken
into account transaction costs and yet others just not good enough so so so yes mostly search and you try to replicated it might not work after you fix some of the however same time once in a while you will come across some ideas that and even after back testing and you know he's still stand up to scrutiny and you know and you traded I've and continue to look and you know we have to have been training one such as that's proved that occasionally research that's important point that I have made my folks is that a
judge that come from the academic research might work that's not exactly as described in the paper so ideas we generate greeting card reform it does not mean that we are just come you know like a slavish copy what we have read always means that we have to make major modification so anything we read in the public domain is to be treated as an inspiration it is not to be treated as the recipe and I made it very clear in michael also that I've quite many strategies and there are occasionally reader would come back couple of
years been working you know what's going on I say well first of all I'm trading that's where you live and continues to work why is it that work for me but not for you the reason is that I make modifications that slightly different from what I wrote a book and I encourage a new reader to do the same and I make sure that they don't copy anybody not meet anybody not even academic researchers in its entirety because obviously very few people will be will be foolish enough to just you know he's 100% of their trade
secrets he's not even there could be souter I can tell you that responded with many attempts and II they are not only just doing research but they may have an interest in the hedge fund or maybe they are trading for some details that they have withheld and so I and i soldi so it is very critical not just save it s written on the net book in the paper but always make sure that you have Modifieds and scrutinizing using your own experience and I qiryat before you trade IKC mentioned in there that you have made
modifications to go strategies overtime what's an example of white making a modification like when you decide if it's a good time to make a modification and I mean is there a difference between just making a modification and dropping that strategy to because it no longer works like how do you know when to make a modification and when to stop trading the strategy well they are occasions where it is quite obvious what is causing those trying to to lose money for example you might be running again in Japan and you might find that the volatility increased
so that you know what our previous with to narrow to be appropriate for the current high for regime change to make your wardrobe and perhaps a look back shorter more responsive to the party or not away might be to imposing minimum volatility snapple just say for sure you follow what's the past North Hollywood you would still be setting it at least with that would be an example of energy to deal with bigger problem and of course this bright idea how to improve their strategy at some future or make some modifications you can also use one
we we must also backed Estes occasionally not just saying that I believe that this place is going to do it well what we have to go back and get to see things now oftentimes when it happens is that it doesn't necessary to return but it might decrease the risk and that might be acceptable just lowering the reason often times he has a great thing you don't have to return sure I K yeah that's that's really great point and while we're on the subject of backtesting I mean let's let's just dive into it so what are
some of the major pitfalls traders can experience when backtesting well a big boy these data snooping bias as we have experienced led many times more simply too complicated and it's just the historical data farewell but it has no predictive power that's usually a problem with respect asking and a cure for that is typically to make sure that the model is simple enough and it actually sensible not just use good result but you know sometimes you have to make sure that sensible is capturing such an environment phenomenon that perhaps well discussed elsewhere so had to restore
confidence that the model is picking up to say no not the noise yeah I mean I guess iconic comes back to what we hit on a little bit earlier about how it's important to understand how the market operates as well as you know the programming knowledge so let me ask you to what extend can aback tests show you the likelihood of a strategy producing similar results in the future like how much do you rely on a back test where we we really got a back tears as some sort of a hypothesis rejection of a number
of words we want to backtest to at least not reduce negative results because it would produce negative pretty much you don't have to spend more time and money on it but she has a duty diesel shop much significance has the highest operational most striking has a high positive operational really means that we are much more confident that mean that but a high operational back has enabled us to high now overheated very quickly you live tree because if you have a shot for example has astronomical shop 10 practically every day every week if not every day
you have a shop age of 10 that's pretty much guarantee that every week and practically every day so if you run destroyed lie just one month you can really tell whether your practice rather because practically every day you available and then you wonderful Monday you find that it has joined on three weeks well media tells you that you're back test is not completely realistic what's your God to factory and transaction costs to your back test so when I say transaction costs you obviously referring to like slippage commissions and market impact that sort of thing how
do you factor that into your back tears well what causes and two more that one need to use data to back as and that's one of my article also why sometimes tried you back testing me quite as high fees because you need to be spread in order to figure out exactly what price are going to execute it just having trade prices that kind of beyond that you have we have to look at the top of full size a lot of this strategy particularly high for instance rival that look very good on paper and particularly favored
by academic researcher you will find out and that you know that your trading just one hundred years but once you intend to execute 10,000 shares sold 10,000 shares this completely gone completely for pot cause you know when you have to take more than a top walls and more money he brings back a crucial work so that's sold price but you also need to know what size and then not to mention even more problematic issues such as you know market impact you know why don't you take the couple liquidity how fast the plan to all of
you could out a limit order there there is a substantial size a factory orderable these issues really in a domain of high-frequency trading even though you may not be a high risk greater discretion and you you you tend to hold for at least a couple of hours this position but if you order into their own a book that is much larger than the typical size you are going to a fact and you are going to have a career and that kind of modeling nowadays being conducted by Hype music and a lot of very complicated mathematical
models coming out of that process but that's the kind of walk much more difficult questions things down some of the common misconceptions around quantitative trading misconceptions well so here's an example let me let me ask you this is quantitative trading just for an elite few who I math wizards what do you honestly believe it's an angle that can be implemented by anyone willing to put in the work I do believe that it is suitable for practically anybody willing to at least learn how to use an Excel spreadsheet I have collaborated with some very successful traders
before that indeed have no programming skills except programming in Excel spreadsheet not naturally they're not doing high fees trading they had doing training that and her wonder day but that doesn't mean it's not gonna take it quantitative trading very slow trading could be could be holding the position false one quarter one year so no one does not need to be math does not need to be a gaming computer with a draining all one needs to do is to be able to gather later and back taxes no matter what would be awesome except ok so trader
who wants to get started in quantitative treanor really wants to dive deeper into the subject and you know give it a real shot where should they start I would suggest that they should stop back trying to backtest some of these simple strategies that they know about all they read about in any sort of trading magazine or block or books and a manual and I have seen that I'm really simple you know why I wear a lot of training books where the strategies extremism such as by I'm gonna be simpler than that but that's a stop
and all of you suggesting that we should always buy when prices drop below 210 be moving out and say something like that that can be quite ready back task and once you gain confidence back as method you might want to make a simple language you know i i i SAT by using six but that doesn't mean best to improve your ability to trade on 22 sort of expand your repertoire quantities once did learning Arabic language then Excel namely and I I always recommend one of three languages at dawn in Matlab so so that's the next
ones you have some confidence back testing using Excel well not to look too so that you can back as wide a variety of especially yeah that's a really great on me and just on the language is there she mentioned MATLAB part on how do you feel about languages such is tried stations AZ language what's the pros and cons of going in the direction of Morgan open source language even though I know you have to do pay for Matlab to use that but Python in both open source languages how do you feel about comparing dogs to
call it a proprietary language like TradeStation season which what are the pros and cons there was a problem languished at a pro-kurdish trading platform why is that you can be very easily time ally trading system so you can back to us and trade life using exactly the same call and generate real-time waters that's very convenient problem is that you want to switch broke all the script written before doesn't work anymore direct brokers easy language we have no use and and not a problem with those languages is that they tend to be broader limiting a on
not design and converse so quickly very extreme example let's say you want to be a new generator I challenge you to do that I don't believe it possible because if you all wanna be with your network from scratch it will probably take months and maybe would because many parts machine learning MATLAB trader or Python traitor or later we would not be doing in your network from scratch we would be using a library that is degrading me in order to generate so that shows you the flexibility of these languages where you know you want to use
this a new life new language that work language well we have to wait until train station development which will thanks for something that I know you've got to published books one is quantitative trading and the second is called algorithmic trading traders expect to learn from reading your two books here well I think particularly answer some of the questions raised only about how to get started it all maybe you never traded before you know maybe someone used it all came from a different business let's say they are trading time job back to football that people really
start on the basics the second both these more often more focus and perhaps into some of the more there and some of the techniques also more complicated seems like common fielder regression models stop or maybe a bit off-putting to complete nor has training but they would be useful to someone who has traded it for a while but would like to learn new techniques so that's what the 2nd ed using something a bit more and more focused and more complicated to the trainers who already have been reading already i cant yeah I mean by size books
on my list to rate and I hope to get stuck into those very soon so I'll link to those in the show notes chat with traders dot com anywhere canisters guard find out more about you and where can i connect with you well I run Apple blocks long thought come and open to comments anybody can come and try to answer them very very rapidly questions I also knew I i have been used to updating the law new article was read and now but now I'm writing you both so I don't have much time to update
its old but these by not an article published every month it doesn't mean that I don't answer comments that's a great idea ottoman on my website and outcome i reset my email and a lot of leaders communicate with me just by so that's why I sometimes when we do ask me questions that I would direct them to the ball because I like to answer questions that many people benefit but other times just want to ask for advice on a particular situation that they have direct and so yeah I know I run courses online and you
know some of the people who have read my books come to disseminate and we would have a real time conversation yea well linked to voice your blog and your website in the shower now it's also a deafening carriage to head on over and check out those links really great resources there earlier also on Twitter I blaze that's right yeah my even my blog I oftentimes my 32 feet and I don't drink very often don't want to you know I love you have any information but I do treat once in awhile when I see out there
so feel free to follow I can and what your hand or change P A N P like a good one and you mentioned you've got a new book coming out when can we expect that well thats should be out well for years who can I just started very recently so and you know you're adding reviewing and then production and marketing tools we probably very CO and what's it about is it a carry on from you lost her yes the topics will be largely different before it will following the second goal in two techniques that I
had not discussed before example machine learning options trading auctions which I have not touched on wall and so far so you will be more advanced techniques and what i've discuss one also however I will just sort of look back to some of the strategies that I discuss and see how they did polls more who died and we will find some that then done either yet that sounds really good and I'm sure many people eager to get their hands on that so many thank you very much for doing this i really appreciate you taking the time
it was it was a lot of fun speaking with you and I mean I shall show a lot details site yet big thanks for doing this and let's speak again so great I enjoy talking to you too goodnight alright also thank you honey you've come to the end of this episode of chat with traders but don't worry more great episodes are on the way to stay updated with each great new episode be sure to subscribe to the podcast in iTunes and we'd love it if you leave us a rating and review will see you next
time on chat with traders
