Skip to content

(Part 4) Time Machine Test – Commodities

May 13, 2010

Results on a commodities basket.

WTI Crude

Natural Gas




As you can see, the algorithm adapts to different classes of commodities and outperforms most of the buy and hold returns. But I want to emphasis that this concept alone is not something I would trade as is (not robust enough). The results are in no way good enough to rely on out-of-sample.

I found that it is much harder for the algorithm to find strategies significant enough to trade on for long periods on these assets. For equities, there is on average 7 different active strategies (i.e. the level of significance is high enough) at the same time. The number is much less with commodities and currencies, the lack of diversification between strategies certainly hurts performance when compared to equity indices. It also add on more exposure to a given strategy adding a lot of volatility to the returns as showed by the numbers above.


  1. May 13, 2010 14:15

    What platform do you test on?

  2. May 13, 2010 14:59

    Wow, drastic differences between currencies / commodities and equities. Any thoughts on why? What instrument did you use to test the currencies / commodities (spot, futures, etc.)?

    • May 13, 2010 15:57

      I used spot data for this test. As for your other question I think there is not a defined follow-through or MR paradigm with these instruments or at least not one this simple algorithm can pick. But this phenomenon needs more investigating.
      Thanks for your comment.

  3. PeteB permalink
    May 13, 2010 15:21

    Hi QF,

    does your algorithm use the 50 long/short strategies as described in David Varadi’s time machine example? If so, those strategies are fairly short-term in nature. That’s fine for equity indices because pre mid-1990’s they exhibited short term price follow through whilst over the past ten years they have exhibited short term price MR. However, perhaps the poorer performance for currencies/commodities is because no short term price tendencies exist for those instruments?

    Perhaps you could include some medium/longer term strategies then your algo will be able to detect longer term price tendencies which may (or may not) exist for currencies/commodities. Just a thought. Keep up the good blogging.


    • May 13, 2010 15:52

      Hi Pete, good observations. I came up with the same preliminary conclusion. I need to dig deeper in the number but eye balling the charts, it seems that the run analysis is rather insignificant on commodities/currencies. It seems that the price tendency for these instruments are much more noisier than equities, making it more difficult to adapt to for the algorithm in its simple form.

  4. charley permalink
    May 13, 2010 15:41

    Agree with PeteB: short term MR and trend persistence (momentum effect) are strongest in equities.

  5. Cornelius permalink
    May 13, 2010 17:35

    Hello QF,

    Thank you for this interesting blog and taking up the topic of “time machine”!

    I would like to ask you to check again the results for equities – the curves are too good to be true.

    I have testest a lot of systems. And if I have learned one thing, it’s that from such an equity curve you can tell something is wrong with the test (even in a “frictionless” calculation).


  6. Paolo permalink
    May 13, 2010 18:45

    Hi QF,

    1. I was wondering how you allocate capital between the signals. Have you used the confidence based approach by MS to weight each signal from active strategies and get an overall weighted signal?

    2. In which way do you believe this approach is superior to a simple equity line trading (i.e. trading strategies above MA only)?

    Great blog and great start btw :-),


    • May 13, 2010 23:01

      Hi Paolo thank you for the questions,

      1. I plan to discuss my method in a future post, I would rather not spoil it for now.

      2. Using MA is another way to go, it is simpler that using say using a Wilcoxon test, thus making it easier to use for many traders. But I do believe that significance testing is more robust than using MAs. MAs are more sensitive to outliers and noise in the data than statistical test (especially non-parametric ones).


Comments are closed.

%d bloggers like this: