Jim Simons is Correct About Trend-Following

In this article I include evidence that was not explicitly presented in a recent TED interview of Jim Simons, where he said that trend-following stopped working long ago. In fact, short-term trend-following was arbitraged out in the 1980s.

Among many other interesting things, Jim Simons, who is a mathematician and a billionaire hedge fund manager, made the following statement in this interview (10:55 minutes from start):

“Trend-following would have been great in the ’60s, and it was sort of OK in the ’70s. By the ’80s, it wasn’t.”

Some trading forum members declared that Jim Simons does not understand trend-following. That was really curious to me. It’s like saying that Niels Bohr did not understand quantum mechanics for which he received the Nobel Prize. If there was a Nobel Prize for trading, Jim Simons would have probably received one.

What is trend-following?

Trend-following means different things to different people but it is important to understand that most trading and investment terms adapt to current market conditions and paradigms: they are as non-stationary as prices. A different kind of trend following was applicable in the 1960s and a different kind is applicable nowadays. Those who applied trend-following in the 1970s chose their models after analyzing the past and those who apply it nowadays choose their models based on different data.

Therefore, the only indubitable claim one can make about trend-following is that it is a trading method with gains that are much larger than losses. Anything else goes into the realms of particular market conditions. Notice that this definition applies even in intraday timeframes. There is no rationale for restricting trend-following in daily or weekly timeframes. However, in this article we will look at daily data because this is what longer-term trend-followers use.

Definition

Trend-following is a trading method that produces an average gain much larger than the average loss

Short-term trend-following

In his TED interview, Jim Simons mentioned how in the past traders would look at the average of the past 20 days and use that as a predictor of the future. That was what trend-following meant to early quant traders.

Let us then consider the following system:

Buy if c – MA(20) > 0
Short if c – Ma(20) < 0

The equity performance of this system in S&P 500 is shown below:

SPX_1-20

The significant outperformance of this system until the early 1980s as compared to buy and hold (before dividends) is clear from the above chart. However, the system stopped working and experienced a large drawdown and by the mid 1980s it turned to a loser.

The second indicator pane shows my Momersion indicator, a measure of momentum and mean-reversion based on 250 daily bars. As expected, the above trend-following system became unprofitable when the equity markets switched from momentum-driven to momersion-driven (a mix of momentum and mean-reversion) in the early 1980s. When markets became mean-reverting in the late 1990s, short-term trend-following was already dead.

Medium-term trend-following

When trend-followers in the 1980s started experiencing large drawdown levels from their short-term models, they increased the moving average period. Next, we consider a 50-day moving average:

Buy if c – MA(50) > 0
Short if c – Ma(50) < 0

The equity performance of this system in S&P 500 is shown below:

SPX_1-50

The longer timeframe worked a little longer but by the mid 1990s any outperformance of buying and hold (before dividends) was also gone. Notice that the system has flat to negative performance after markets turned mean-reverting in the late 1990s.

Longer-term trend-following

By the early 1990s, some traders thought that the problem was the moving average period and they increased it even more to 200. Next we consider the following system:

Buy if c – MA(200) > 0
Short if c – Ma(200) < 0

The equity performance of this system in S&P 500 is shown below:

SPX_1-200

The system with the 200-day moving average provided a decent predictor of future price action until the late 1990s. Afterwards, the system performance is erratic with large drawdown levels of the kind that the early traders of the 1970s would not accept.

Curve-fitted/naive trend-following

When traders realized the moving average was no longer a good predictor of price, they tried two moving averages, a system known as a moving average crossover. Note that these systems are curve-fitted and ahve no intelligence in pairing market returns with system returns. As a result, they can fail at any time. Examples are offered in my new book, Fooled by Technical Analysis and in this article. Traders who use such systems to trade are basically gamblers.

Let us consider the following curve-fitted system:

Buy if MA(50) – MA(200) > 0
Short if MA(50) – Ma(200) < 0

The equity performance of this system in S&P 500 is shown below:

SPX_50-200

The significance underperformance of the buy and hold (without dividends) after the mid 1980s is a strong indication that this system is random but this can also be shown with statistical analysis. Basically, the fitted system represents a bet that the markets will continue uptrend while any correction will not be followed by extended sideways action. If these conditions change, the losses will be large and even cause total ruin.

Price series momentum was gradually arbitraged out starting in the early 1980s and continuing into the 1990s. The main reason for this is simple: too many traders were using these models and it was not possible for all of them to profit. Some had to lose and in the process the systems became unprofitable.

Examples from currencies

Short-term trend-following in GBPUSD

GBPUSD_1-20

It may be seen that the 20-day average stopped being a good predictor of future prices in the early 1990s. Afterwards, the associated system is a loser.  The 100-day moving average also stopped at the same time although it did not offer as smooth of an equity before that, as shown on the chart below

GBPUSD_1-100

I started trading in the late 1980 after graduating from university. I already knew that short-term trend-following no longer works and I used a short-term crossover of the following type:

Buy if MA(5) – MA(25) > 0
Short if MA(5) – Ma(25) < 0

The equity performance of this system in GBPUSD is shown below:

GBPUSD_5-25

 

Jim Simons is probably right that the “easy” trend-following is dead long time ago.

If you found this article interesting, I invite you follow this blog via any of these methods: RSS or Email, or follow us on Twitter

If you have any questions or comments, happy to connect on Twitter: @mikeharrisNY

Charting and backtesting program: Amibroker

Disclaimer

Technical and quantitative analysis of Dow-30 stocks and 30 popular ETFs is included in our Weekly Premium Report. Market signals for longer-term traders are offered by our premium Market Signals service. Mean-reversion signals for short-term SPY traders are provided in our Mean Reversion report.

Copyright Notice

This entry was posted in Technical Analysis, Trend following and tagged , , , , . Bookmark the permalink.

19 Responses to Jim Simons is Correct About Trend-Following

  1. Bo says:

    Mike,

    Another great article which leads me to do some testing on my side. I tested the naive 50/200 crossover system with slight modifications: long when 50dma>200dma+3*atr;
    short when 50dma<200dma-3*atr;
    do nothing in between.
    I use a five day nearest neighbour as stop and risk 80bps per trade. The system trades 138 assets(equity index future/bond/currency/commodity) all over the world. The testing period is from 1/1/2007 till 9/21/2015. While the backtests duration is not that long, we have boons then bust then boon again and we have a series of macro crisis. I think this is a fair representation of what might happen in the future. Here is what I found:
    1. On an individual asset basis, the results are largely random. ExAnte there is no way to tell which one it will work which it will not work.
    It works very well on Brent Crude(+100% vs -30%); very poorly on Gasonline(-52% vs +80%);
    Silver(+67%vs +3%); Copper(-40% vs -20%);
    Sugar(-80% vs +70%);Coffee(+18% vs -30%)
    GBPUSD(+13% vs-13%);GBPJPY(-20% vs +8%)
    SP500(+44% vs 38%); R2000(-3% vs +48%)

    2. On a portfolio level, the annualized return is 39% the sharpe ratio is only 0.35. In 2012 it suffers a nerve breaking loss of -75%.

    My conclusion is: the trend following system doesn't work on individual asset level. It can work on a portfolio level. You need to apply it on a risk parity basis to a large number of assets. This is the only way that you could catch some of the exceptional trend that gives you extreme large returns to cover the many small losses.
    Nothing is easy but it can work.

    • Hi Bo,

      Great work! I wish you could post the findings in a blog.

      On a portfolio level, trend-following may work if the "spoilers" are not included from start. The t-statistic for a Sharpe ratio of 0.35 and for a period of 8 years is equal to 0.99, meaning that results are not significant since the value is not even one standard deviation away from the null of zero return.

      In order to get a t-statistic of about 2, you will need to have either about 32 years of results at same Sharpe or a Sharpe of 0.7 for the 8 years (of course without any curve-fitting)

      Best.

  2. Mike R. says:

    Guess you should tell David Harding to shut down his $25bln fund he built since the last '90s since trend following no longer works.

    • Dear Mike,

      I just ask you kindly to actually read the blog more carefully and realize that the claim by Jim Simons, which I support, is that short-term trend following does not work, as it is demonstrated with analysis. Some people think that short-term trading following is the only viable approach. Some others disagree.

      Longer-term trend-following works, as it is also demonstrated in the blog, but at higher risk levels than in the 1970s.

      As far as specific funds, one cannot know their exact method and/or edge. Trend-following means different things to different people as it was made explicit in the blog.

      • Mike R. says:

        "Many traders and funds that are using trend-following that is not the pure, short-term type, Jim Simons talked about in his interview mentioned before, are risking losing a low of money or even getting completely ruined."

        So you're saying if these funds are NOT using pure, short-term trend following then they risk losing a lot of money because I thought the argument was that they would be risking losing a lot of money if they WERE using pure, short-term trend following situations.

        I understand your point was that short-term trend following doesn't work anymore but you said trend following means many different things to many different people yet in your conclusion you stated, "Jim Simons is right that trend-following is dead long time ago and the non-believers should only question their own experience and knowledge of the markets." I believe you should clarify that a little better with exactly what "type" of trend following you're singling out here.

        Appreciate your work though.

        • Hello Mike,

          Trend-following in the sense of quants (and Jim Simons) is that a reasonable short-term period of average prices is a good predictor of future prices. For example, in the TED interview Jim Simons referred to an average of 20 days.

          This type of trend-following no longer works because markets are now mean-reverting. As a result, trend-followers now have increased the look-back period hoping that the structural bias of markets stays in place.

          However, this longer-term trend-following invites more risk. It is a fact that there are still many trend-followers around but this style of trading is more risky. This is the main point and I think it was made clear in the blog:

          "Obviously, if equity markets continue to deliver premium with no extended whipsaw, the naive models that are used nowadays may stay profitable but that is gambling. "

          • Nick Flyte says:

            Hello Mike,

            I found this part of your comment rather interesting: "this longer-term trend-following invites more risk".

            Risk is generally associated with volatility, which in turn generally benefits outlier approaches such as trend-following. So, you're saying that what we conventionally equate with a low-risk environment is actually a high-risk environment for trend-following strategies?

            Of course, volatility is only really a simple proxy for risk (and one which I would suggest has originated in long only equity management), and I do share your (and Simmons's) view that markets have become more mean reverting in shorter time-frames.

  3. Bo says:

    I ran another test with short term naive trend following system. This time we use 5 day-10 day crossover system. The annulaized return is 14% while sharpe ratio improved to 0.5. On assets level, it does very poorly on US equity markets but performs quite well on certain commodities and currencies. Againt the signals are all random on individual assets. It seems to me the exit rules, position sizing and the capability to trade as many assets as possible are far more important than entry signals.

    • Hi Bo,

      You may be right but there is risk of data-snooping bias here. You have already done several tests. This has caused the data-mining bias to increase. You are now forming hypotheses based on what you already know from previous tests and the data. This means that the t-statistic must be corrected for the multiple tests. A Sharpe ratio of 0.5 is not enough to account for that (See Benferroni correction)

      "…but performs quite well on certain commodities and currencies"

      This increases the risk of selection bias.

      In addition, I tend to disagree than trading many assets is more important than trading signals. Some trend followers learned that the hard way in 2008 when all assets were correlated.

      As you said: "Nothing is easy…"

      Have you included friction in the tests?

  4. Hello Nick,

    "So, you're saying that what we conventionally equate with a low-risk environment is actually a high-risk environment for trend-following strategies?"

    No, I am saying that the higher the lag of the predictor (ex. 200 days vs. 20 days), the higher the risk of a larger drawdown. Price often takes longer to cross a 200-day moving average than a 20-day. That is all.

    Best.

  5. John Butters says:

    Thanks for a very interesting post.

    I am not sure what to do with it. It's all about short-term trend following but using a 200-day moving average surely takes us into "long-term" territory. It looks like this no longer works on the S&P 500. Trend followers do make money trading a range of markets, though, as we can see from the Newedge Trend Index of hedge funds.

    http://www.newedge.com/en/newedge-indices/

    Also, markets maintain their trendiness — one can see that in academic indicators like TSMOM, in signal-to-noise ratios and in simple trend-following systems like the one in Greyserman & Kaminski's text book on the subject. I'm not objecting to your article as I know you said it was about short-term trend following. I suppose the interesting question for me is whether I should react to Jim Simons' comments by removing systematic trend following funds from my asset allocations — it seems to me not, since they have all gone medium-to-long term where the strategy appears to work consistently.

    • Hello John,

      Thank you for your input. Presenting a counter-argument in a constructive way is always good and as a matter of fact absolutely necessary for progress.

      I agree that most CTAs and other systematic trend-followers make money trading a range of markets but there are reports that some of them also employ short-term strategies, or even HFT, to compensate for deteriorating performance.

      In the last 12 years, other in 2008 and 2014 (equity market related gains since most concentrated on that), systematic traders are not doing very well, according to this index:

      http://www.barclayhedge.com/research/indices/cta/sub/sys.html

      The idea is this: if markets go in an extended period of whipsaw, such as the emerging markets for example (look at EEM), then longer-term trend-followers may register losses depending on their risk management method.

      But yes, this was about short-term trend-following, the type that makes sense to quants. AFAIK, longer-term trend-following gains are mostly due to luck, and as you said, diversification.

      Again, thanks for your input.

      Best.

      • John Butters says:

        Dear Mike,

        Thank you for this and for the link to the paper below.

        Indeed, trend followers have struggled recently. The NewEdge trend indicator (which is not an index of actual returns but a theoretical series) shows that there has been a lack of trendiness in markets since 2011. Also, one would expect lower returns from a futures based strategy at a time of low interest rates (because they earn a return on cash).

        But it does seem a stretch to put the prior performance of trend following entirely down to luck. I suppose one could say that it is luck that markets have had enough trends in the past to make the strategy profitable, but one could also take it as evidence that trendiness is a persistent feature of markets. I think of it as being related to Paul Samuelson's suggestion that markets are "micro efficient" but "macro inefficient": I theorise that trends occur when macro inefficiency corrects itself, something that seems to happen from time to time.

        Best wishes,

        John

  6. John Butters says:

    Incidentally, just another thought, but the performance of momentum strategies does not seem to have got worse, according to this paper (nice table on page 51):

    http://www.valuewalk.com/wp-content/uploads/2015/06/SSRN-id2607730.pdf

    (PDF)

    • Thanks but these studies are like medical research or even worse. There are counterarguments:

      http://papers.ssrn.com/sol3/papers.cfm?abstract_id=968176

      Everything on hindsight is easy. But will the future be the same as the past? If the samples of these studies are representative of the population then there is hope. If not, surprises are in store.

      Best.

  7. Simon says:

    So trend following doesnt work since the 1980s ?

    Am I to assume there have been no trends in markets since the 1980s ?

    • The article is not about trends. It is about whether past prices can predict future prices and especially at low lag. The trends you see are only there on hindsight. Trading requires ex ante analysis. These are two different things. People who do not trade may not understand the difference.

  8. Rick says:

    This is what someone wrote in his blog [comment section] about this post. I think it is close to a libelous accusation.

    "the gist of his post is that trend following of any kind is going to lose you a lot of money and is the work of the devil"

    I could not see such conclusion from reading your post. If you need the link to the other blog I will email it to you.

Leave a Reply