Sunday, October 4, 2015

Carver, Systematic Trading

The days of Richard Dennis and his “turtles” with their alleged 100% per year profit are long gone, but their mystique lives on. And with it comes one attempt after the other to emulate them, to create trading systems that will knock the socks off the competition.

Robert Carver is more modest—and more realistic. At the same time he has more to offer the investor or trader who has a spark of creativity and intellectual curiosity. Systematic Trading: A Unique New Method for Designing Trading and Investing Systems (Harriman House, 2015) is a thoughtful, and thought-provoking, journey through the process of creating modular rule-based portfolios.

Although the book addresses three classes of traders and investors—the staunch systems trader, the semi-automatic trader, and the asset allocating investor, Carver is at heart a systems guy. He himself runs a futures trading system with around 45 instruments, eight trading rules drawn from four different styles, and 30 trading rule variations. But this doesn’t mean that he is writing only for those with large portfolios who can code. It does mean, however, that his book will be of value only to those who either already think systematically or are open-minded about learning how to analyze and assess the ingredients of a model investing framework. I would wager to say that this group should include every investor and trader, though in practice of course it encompasses but a tiny fraction of people who have money in the financial markets.

Here I’m going to be decidedly unsystematic and pluck out two ideas that are illustrative of the topics covered in the book.

First, according to Carver, the most overlooked characteristic of a strategy is the expected skew of its returns. “Assuming they have the same Sharpe ratio, the returns from a positively skewed asset will contain more losing days than for those of a negatively skewed asset. But the losing days will be relatively small in magnitude. A negatively skewed asset will have fewer down days, but the losses on those days will be larger.” (pp. 44-45) Equities normally have a mildly negative skew, foreign exchange carry is a negative skew strategy (sometimes disastrously so—think the Swiss franc in January 2015), and gold tends to have a positive skew. Trend following strategies and long option strategies have a positive skew; fixed income relative value (remember LTCM?) and short option strategies have a negative skew. VIX futures have a highly positive skew, “around four times higher than their underlying index.” (p. 46) Negative skew trades often seem more attractive; after all, they are like selling an insurance policy. But managing risk in these trades is more difficult since losses are large and infrequent. Moreover, they often require leverage to achieve decent returns in normal times, so they get killed in bad times.

Second, systematic trading requires forecasting. “A forecast is an estimate of how much a particular instrument’s price will change, given a particular trading rule variation.” (p. 102) “A forecast shouldn’t be binary—buy or sell—but should be scaled. … There are three reasons why scaled forecasts make sense. Firstly, if you were to examine the returns made by a trading rule given the size of its forecasts, you’d normally find that forecasts closer to zero aren’t as profitable as those further away. Secondly, binary systems cost more to trade, since to go from long to short you’d need to sell twice a full size position immediately. Finally, the rest of the framework assumes that the forecasts you get are not binary or lumpy in other ways. It’s better to see forecasts changing continuously rather than jumping around.” (p. 113)

To set forecasts, Carver recommends using volatility standardization. Forecasts are “proportional to expected risk adjusted returns. For example, suppose that the Bund has expected returns of 2% a year and an expected annualized standard deviation of 8%. Schatz futures have an expected return of 1% a year, but you only expect volatility of 2% a year. After adjusting for risk the expected return on Schatz … is twice as much as on Bunds…. That implies the forecast for Schatz should be twice the forecast for Bunds. … If you continuously adjust your estimate of expected volatility then you also cope with risk changing over time.” (pp. 114-15)

Carver spent ten years in the City of London—initially trading exotic derivative products for Barclays and then serving as a portfolio manager for the hedge fund AHL, where he created its fundamental global macro strategy and managed its multi-billion dollar fixed income portfolio before retiring from the industry in 2013. So he isn’t just some ordinary Joe with a computer and a bunch of back-testing software. He has clearly thought about what makes a good systematic trader and a good systematically-driven portfolio. We can be grateful that he decided to share his insights with us.

9 comments:

  1. Thank you. I have also worked for this industry. You have to also consider the possibility that Carver confuses the skew due to curve-fitting with a positive skew from the returns of an asset class. Trend-following rules are usually curve-fitted to past data. It is peculiar to compare asset classes to trading rules. This shows confusion. trend-following is not an asset class.

    "Carver spent ten years in the City of London—initially trading exotic derivative products for Barclays and then serving as a portfolio manager for the hedge fund AHL, where he created its fundamental global macro strategy and managed its multi-billion dollar fixed income portfolio before retiring from the industry in 2013."

    Carver looks like in this 40s or early 50s. If his models were successful and his understanding was sold he would be in high demand by the fund industry. The fact that he "retired" raises some issues.

    and excuse me on this:

    [That implies the forecast for Schatz should be twice the forecast for Bunds]

    This is an incomprehensible statement. Do you mean position size?

    Thank you

    ReplyDelete
  2. Nope, no confusion at all. I'm very clear about the distinction between skew of trading rules and asset classes, as you would know if you had actually read the book (or the paragraph that is quoted).

    In my book I spend a great deal of time discussing how to avoid curve fitting.

    In any case I'm not sure how fitting affects skew; normally you fit to maximise sharpe ratio, although you could fit to maximise skew it's unusual. What affects your return profile is the combination of:

    - the style of trading rule you are using.
    - the underlying asset you are trading

    If you're curious I'm 41 (thanks for the comment about 'early 50's'). You might consider the following:

    a) you need to have been paid a reasonable amount to retire at 39 (particularly paying UK tax marginal rates of nearly 50%), which implies I must have been kind of okay at my job
    b) I must be quite good at trading to be living off the reasonable amount of money. Good enough that I expect to make a lot less from book sales than investment income. I don't sell software, sell newsletters, or do 'trader education'.
    c) not everyone wants to work until they are dead. Particularly if they apparently already look 10 years older than they really are. Some of us would like to spend more time with our families and generally enjoying life; and perhaps writing the occasional book.

    A forecast is something that is proportional to sharpe ratio (mu / sigma). As you probably know position is proportional to mu / sigma^2. Your position would be scale x forecast / sigma, where sigma is the standard deviation of the instrument and scale depends on various other things like how much capital you have and the tick value of the contract. Again if you read the book this might make more sense to you. You can read a preview for free from the publisher.

    ReplyDelete
    Replies
    1. [As you probably know position is proportional to mu / sigma^2. ]

      You probably refer to continuous Kelly ratio, as explained in the book by Dr. E. Chan. This is a leverage factor. I have never heard that

      [A forecast is something that is proportional to sharpe ratio (mu / sigma)]

      Please provide a reference for the above statement other than your book. You may be right but I believe you are wrong and like many others you have misinterpreted the material in Dr. Chan's book.

      Sharpe ratio for SPY is about 0.6 and sigma is about 0.16 (haven't checked latest numbers) Are you suggesting to size positions based on a leverage of 3.75 or about?

      If you do and I correctly interpret your assertions, I must ask you if I should take you seriously.

      Delete
    2. No position *proportional* to mu / sigma ^2 is just Markowitz. Nothing to do with Kelly. This is true regardless of your appetite for risk or your portfolio size.

      (Note I didn't say position is EQUAL to mu / sigma ^2. It would be helpful if you actually read the book, or the review, before we have these debates)

      (By the way I've not read Chan, though I'm sure it's very good. Nor do I use the term leverage factor)

      "[A forecast is something that is proportional to sharpe ratio (mu / sigma)]
      Please provide a reference for the above statement other than your book. "

      Reference to this is: My book. I made it up. Since it's my definition I think I'm allowed to do that. As long as it produces correct positions (and it does, as you'd know if you read chapters 9,10 and 11) then I think I'm allowed to do that.

      "Sharpe ratio for SPY is about 0.6 and sigma is about 0.16 (haven't checked latest numbers) Are you suggesting to size positions based on a leverage of 3.75 or about? "

      No, nothing of the sort. To decide what your position is you'd have to take a whole lot of other stuff into account, portfolio composition, risk appetite and correlations.

      PS
      If I were you I wouldn't expect a SR for SPY of 0.6. Chapter 2. If you're ever going to read the book.

      Delete
    3. (different anonymous here...)
      Hi rob,

      I like your writing and i'll be buying the book.
      I'm also a bit confused about the whole 'scaled forecast' and position sizing thing.

      It makes sense to have your ultimate position in instrument X dependent on the volatility of instrument X (ie. low vol instrument = higher position size). Assuming your forecasts are instrument independent (ie. your trend following rule doesnt actually forecast an increase in price of Y), are you saying that you take this into account in your forecast such that your scaled forecasts aren't comparable between instruments?

      Cheers

      Delete
    4. Hi 'different anonymous' sorry for the delay. I don't follow comments on this blog religously.

      Yes the ultimate position of X should depend on the vol of X. So there is a an extra step where I take the scaled forecast (instrument independent, agnostic about vol) and convert that to a position. One of the things that happens in that step is that we divide by volatility.

      Delete
  3. Great answer Rob, you just sold one more book!

    ReplyDelete
  4. Forgot to add; thank you very much Brenda for the great review. And thank you Hakan.

    I'll leave the readers of this blog to decide if they will take comments by an anonymous user into account (whose motives and incentives might not be completely pure).

    ReplyDelete
  5. Thanks for putting a great book together Rob. I have to say your sense of humour in dealing is trolls is commendable!

    ReplyDelete