A number of the commodities included in the testing portfolio don’t have an active underlying options market so I’ll be using historical volatility over a period similar to that used to measure the skew. Thanks, I was beginning to think that I may well be the only one reading this With that in mind if anyone has any further comments/criticisms of either the research paper or anything that I’ve written, then please feel free to post prior to my carrying out the test.
The final issue I have prior to carrying out the back-test is one that applies to any rotational strategy … how frequently should we rebalance? How important this issue is should really depend on two factors. Firstly how many candidate markets are included in the overall portfolio and secondly, how frequently the ranking measure – in this case, skewness – changes the position of each market in the ranking. Of course the more frequent the rebalancing the greater the impact of transaction costs will be on the strategy’s profitability. In this case the authors choose to rebalance monthly. This seems reasonable due to the small number of markets as well as the long lookback periods used to measure skew which will both result in relatively slow changes to the individual market rankings. That said I’ve never been a big fan of rotational strategies that result in once a month wholesale changes to the overall portfolio. My preference has always been to scale in and out of positions – a type of dollar cost averaging – so in addition to monthly I will also be testing a daily ranking that will make incremental position changes rather than all in or all out positions.
Results time … Below is a chart from the authors’ research showing the performance of their Skewness portfolio vs a few other commodity strategies that the authors have previously researched. At first glance it looks pretty smart, right? Nice sloping upward curve with no big drawdowns. And here are my results over the same period, same monthly rebalance and same 20% threshold but applied across a more diverse group of commodities … … not so smart And here over a longer time period, 1971 – 2016 … … even less smart. I also looked at a few other options by altering both the lookback period used to measure the skew as well as the entry threshold. This of course is a classic case of back-test overfitting, but just for fun here is the best I could come up with using a 500 day lookback and only trading a combination of long one market with the lowest skew and short one market with the highest skew, rebalancing monthly … Actually in this case the results over the last 20 years don’t look too bad, however the prior 20 years are indeed too bad. One of the factors that was important to me in carrying out this exercise was to test for skew predictability of future returns across as diverse a range of liquid and tradeable commodities as possible. My thinking was that this would help in establishing the robustness or otherwise of the trading strategy. If the strategy had then looked plausible – which it clearly doesn’t - it was next my intention to check for any particular market outliers. If you take the results using the 20% threshold over the longer 45 year test period, here they are, market by market … So even with the underwhelming results at a portfolio level, if we were just to remove one market – UK Natural Gas – our outcome over 45 years would be truly unimpressive. What’s more the stellar results from UK Natural Gas derived significantly from one hugely profitable trade by going short in March 2005 and exiting some 4 years later. In fact while we are on the subject of skew you can see this single UK Natural Gas trade at the far right of the following trade distribution plot. So my conclusion … yes, historical skew is indeed a strong predictor of future returns … if you exclusively trade UK Natural Gas
Thanks I appreciate that. As to why I arrived at such a different outcome from the referenced research paper, I’ve already hinted at a few reasons in my earlier posts, however in summary I’d suggest the following key issues … · If you wish to examine the relationship between characteristics or risk factors and subsequent returns in commodity futures markets it’s essential that you sample this across a diverse a group of commodities as possible rather than testing a portfolio which is limited to USD denominated markets that include commodities that are almost replicas of each other when it comes to their return profile. · 26 years of daily data may sound like a lot …. but 45 years produces quite a different outcome · It’s important that all commodities within the testing portfolio can potentially have an equal impact on the strategy’s profitability or otherwise and so weighting allocations should be based on each market’s recent volatility rather than nominal dollar values. · As it’s unclear exactly which methodology the authors used to construct a continuous price series for each market, it’s possible that they have not accounted for the price gaps when rolling from one contract to another. I think it’s unlikely given the authors’ combined experience in researching commodity futures, however if so, then that would certainly help to account for the large discrepancy between our results. · The paper doesn’t disclose individual market contributions to the overall portfolio results. This is concerning as it’s possible that some of the “non-tradeable” markets such as Pork Bellies or Electricity may be the major contributors to the strategy’s profitability in the same way that UK Natural Gas dominated my own results. As an aside I’m certainly not suggesting that the use of the skew of a market’s historical returns should not be employed within strategy development, however using it as the sole ranking measure in a rotational strategy designed to trade commodities in my opinion has no value. That said, I’d certainly be interested to hear from others as to their thoughts on the use of this relatively simple statistical measure within their strategy design process. Meantime I’m also going to email the paper’s authors to see if they would like to offer any further thoughts as a result of the above findings.
Well, in that case, the point of this whole thread was to examine the sort of skew which stays, rather than comes and goes. At least that was the claim made by the authors of the paper the OP mentioned (27 years worth of data).