Better System Trader

Better System Trader

Better System Trader is the podcast and blog dedicated to systematic traders, providing practical tips from trading experts around the world.

012 – Ernest Chan

Play in New Window|Download

Inter

Ernie Chan is an expert in the application of statistical models and software for trading currencies, futures, and stocks. He has built and traded numerous quantitative models for investment banks and hedge funds.

He is now the Managing Member of QTS Capital Management, a commodity pool operator and trading advisor, managing a hedge fund as well as individual accounts.

Ernie is the author of 2 books, maintains a popular quantitative trading blog and teaches courses and workshops in trading and finance.

Today we talk about many aspects of quantitative trading, including how market crises impact momentum strategies and how to manage the impacts, when to use stop-losses and when they don’t make sense, automating trading, managing funds in a portfolio of strategies and a simple money management approach which aims to limit drawdowns while maximising returns.

‘A lot of my ideas are just based on logic’ – Ernest Chan

Click To Tweet

Topics discussed

  • Where to find trading ideas
  • The first aspect of a market to identify before building a strategy for it
  • Momentum crashes and the performance of momentum strategies after a financial crisis
  • How to manage the times when momentum strategies aren’t working
  • When stop losses should be used and when they don’t make sense
  • How to limit drawdowns while maximising growth
  • Factors to consider when automating your trading
  • How independent traders can avoid competing with the big trading firms
  • When you need to worry about market microstructure and when it doesn’t matter
  • Managing funds for a multi-strategy portfolio
  • The hardest part of trading

Resources & Links mentioned in this episode

Quotes

Stick to the stuff that actually generates profit with limited risk.

Click To Tweet

Always prepare, think about the worst-case scenario.

Click To Tweet

The biggest lesson is not to be overconfident with a strategy.

Click To Tweet

Trading is a tough business, do not expect to succeed in a short time.

Click To Tweet

Top tips from this episode

  • Simple stuff works – Ernie has worked for some big name investment companies that had enormous resources, perhaps unlimited resources. In those environments he was exposed to the greatest minds and work but when he went out on his own he chose the simple stuff. He wasn’t out to impress anyone with complicated strategies, it was all about what worked, what made money and that was the simple stuff. I’m sure we’ve all been guilty in the past of dismissing a strategy because it seemed too simple or perhaps we’ve overcomplicated a strategy, I know I’ve done both but it’s a good idea to keep simplicity in mind.
  • I really liked the concept of Constant Proportion Portfolio Insurance, which aims to limit drawdowns and maximise growth under the constraints of the drawdown level. If you’re in a situation or you have a preference to limit drawdowns to a specific level, this could be a great strategy to employ.
  • I also found his comments on Momentum crashes interesting. Momentum strategies can take a while to recover after a financial crisis and he suggested one way to manage that was by trading mean reversion strategies as well.

Transcript

Got A Question, Topic or Guest you want to see on the Podcast?

Do you have a specific question, topic or guest you’d like to see on a future podcast episode?

Click here and submit your suggestion for a chance to have it featured on an upcoming podcast episode.

Subscribe to Better System Trader and never miss another episode!

subscribe-with-itunes-button stitcher soundcloud android

Please support the podcast by giving an honest Rating/Review for the show on iTunes!

Episode Released:

21 June 2015

Related Posts

Comments

  1. Hello Andrew,

    Good interview you had with Dr. Chan. He is definitively an expert on algo trading. Despite some recent articles in the blogosphere that question the use of stop-loss, I think his suggestions along these lines were useful. I just wanted to clarify his reference to Kelly In the interview (~23:00). You mentioned the Kelly ratio, also known as Kelly fraction. However, he works with the Kelly leverage. These two are different notions. The Kelly fraction is for position size determination in discrete trading, a fairly dubious concept, and the Kelly leverage is for leverage determination in the case of fully invested equity.

    In ~23:13 Dr. Chan defined the Kelly leverage as the “average return squared divided by the standard deviation of returns”. I think he obviously had something else in mind and he made a little error. The Kelly leverage is the ratio of average return divided by the variance of returns. Using this leverage results in optimal equity growth, at least theoretically and assuming continuous rebalancing. However, this optimal leverage depends on the look-back period used to calculate the mean return and variance and usually results in excessive risk that can wipe out trading capital due to a single bad trade. Hence, the need for a stop-loss to prevent that from happening. Regardless, I just wanted to clarify that you were not talking about the same thing and also point to this error in the definition. Otherwise, I think this was a valuable interview and traders should pay attention both to what Dr. Chan said but also to things he did not speak of directly but alluded to.

    • Hi Michael, you’re correct, Ernie WAS talking about Kelly leverage ~23:00 however I misunderstood him to mean Kelly ratio. Thanks for raising it.

      Regards,
      Andrew.

    • Hi Michael,
      You are right that I was speaking of Kelly leverage. You are also right to point out that it is equal to average returns divided by the variance of returns, which I mistakenly said was the standard deviation of returns.
      Thanks,,
      Ernie

      • Hi Ernie,

        What is your opinion about machine learning? Have you tried developing any system with it? I understand you are an advocate of simple, logical rules and that avoids some of the data-mining bias issues faced by machine learning. I am too.

        @Andrew

        There is a huge difference between how retail traders and pros like Dr. Chan approach trading. Pros must use cumulative and often daily stops due to company policy (ex. forex dealing desks). Some funds do not allow machine learning due to data-mining bias issues and developers must come up with logical ideas that are then tested for significance in real trading Backtested results have little value in professional environments. One reason that retail traders suffer losses is because they place too much emphasis on backtests. There are many retail traders who post results of patterns with 30 or 40 trades in 10 or 20 years and think these results mean something when in fact they mean nothing without detailed significance tests on unused data to avoid data-snooping. Again, great interview with Dr. Chan. I always pay attention to what he has to say.Thanks.

        • Hi Michael, I agree. Alot can be learnt by understanding how the pros approach trading and that was one of the main reasons why I started the podcast, there is always more to learn!

          Glad you enjoyed the interview with Ernie.

          Regards.

        • Hi Michael,
          We don’t use much machine learning in our trading, though I have done quite a bit of work researching such methods for trading. The reason is as you suggested: there are too many parameters for fitting in a typical machine learning system, leading to data snooping bias and miserable performance in a walk-forward test.

          But we do partner with professionals who are able to make this method work, and I do teach these methods in my online webinars (epchan.com/workshops) for those who are interested in trying them out.

          Ernie

  2. Hi

    Great podcast, Andrew and Ernie.

    In regard to market regime, I have a question. Is it possible to develop some kind of technique, filter or indicator or something, to judge quantitatively when the market is trending and exhibiting positive serial correlation, and when it is not or rather exhibiting negative serial correlation and mean reverting? What would this technique be?
    The following blog posts kind of discuss the state of the market, US market, and how since about 2000 the markets have been mean reverting. That is, buying strength would be a losing proposition. Of course, the posts also show that before 2000 the markets were trending so buying strength would have been a good strategy.
    https://marketsci.wordpress.com/2009/01/05/testing-tm-rule-1-buy-new-lows-not-new-highs/
    https://marketsci.wordpress.com/2009/01/05/testing-tm-rule-2-buy-the-market-after-it%e2%80%99s-dropped-not-after-it%e2%80%99s-risen/
    I think such a technique might be used for market timing, a kind of filter of sorts. Something to be tested with that hypothesis in mind, anyway.

    I’ve tried using the Durbin-Watson statistic but didnt get much joy from it. There are a number of possible alternatives: Hurst exponent, Perry Kaufman’s efficiency ratio (measures level of noise), Lo’s variance ratio, DV2, TSI, or even something like the above in the links – ten day high or something? Would any of these be useful?

    Next, in regard to monte carlo testing while keeping the market regime in mind, how could you go about this for backtests? For example, while the market regime is exhibiting positive serial correlation you wouldnt really want to completely randomize all trades as that would break the correlation and lead to misleading results, however how about when the market is exhibiting negative serial correlation, could you randomize trades in this environment?

    Looking forward to reading your reply.

    Thanks for the great podcasts.

    Shawn

    • Hi Shawn,
      I don’t think that many regime detection techniques work very well, especially techniques such as Hidden Markov Models. One of the reason is that regime switching happen not frequently enough to allow good statistics for testing.

      I also don’t think that a Monte Carlo simulation will let you test regime switching detectors, since much of the higher order correlations in the market will be missing from a Monte Carlo simulation.

      The only indicator that makes sense to me is order flow. In a shorter time frame, order flow is a good predictor for trends. So you can expect an absence of strong order flow to be a predictor of mean-reversion.

      Instead of regime detection, there are a few indicators that may be good for “crisis” detection. For e.g. VIX, the credit spread, etc.

      Ernie

      • Hi Ernie

        Thanks for the reply. That’s good to know about regime detection techniques. I suppose these techniques will always be lagging too.

        I’m not too clear about what order flow is, but I just bought your newer book and see that you explain that in there. So when I get to that point, I’ll read about it then.

        Thanks again

        Shawn

  3. Thanks a lot, great and very interesting interview.

    Is there a way to download it?

    Thanks a again.

    • Hi Socrates-Trader, apologies for the delayed response, your message got stuck in our spam filter.

      There is a download link just below the play button.

      Regards,
      Andrew.

  4. You mentioned a research paper stating that after 30s crash momentum did not work for decades. Can you please make a reference to author/paper. What was in portfolio? Thank you.

    Stefan

Trackbacks

  1. […] Interview with Ernest Chan [Better System Trader] Ernie Chan is an expert in the application of statistical models and software for trading currencies, futures, and stocks. He has built and traded numerous quantitative models for investment banks and hedge funds. He is now the Managing Member of QTS Capital Management, commodity pool operator and trading advisor, managing a hedge fund as well as individual accounts. […]