Traders' Roundtable

I use CSI data. I have noticed that profitability doubles in various markets from Rough Rice to Crude Oil to Natural Gas (among others) by rolling the contracts 90-120 days before expiration, as opposed to rolling them based on open interest and volume. (I'm using a long-term trend following strategy) I have a hard time believing that this would be the case in a live portfolio, but I cannot seem to find a bug that would explain it away. Has anyone else had this experience? Does anyone have any idea how to verify the validity of the results other than trading the system in real time and comparing it to the simulated results?

Many thanks to anyone who participates in this discussion!!

-ColdFact

You need to understand the term structure for each futures contract in your portfolio as well as if it is in contango or backwardation. In general under some circumstances upto 1/3 of commodity returns are determined by the term structure vs the trend. For some of the commodities you mentioned, energy complex, being 90 days out is not a big issue but for some the markets could get thin.

I'm confused as to how backwardation or contango is relevant. I have noticed that there is very little difference (and it seems random as to whether the difference is positive or negative) for the grains/dairy/etc - but the difference in return is hugely positive across every single energy market. What I'm interested in knowing, is what could make this be the case? I have tested it in every 5 year period since 1973, and it is consistently true that using a rolling method 4 months-ish out makes the profitability much higher - generally as a result of having a higher percentage of wins.

I believe the reason this happens is that the "price insurance" premium embedded in the futures prices decreases as time to delivery decreases. By rolling earlier your premium is not reducing at the rate it would as delivery becomes closer. I would think this would be observed on long contango contracts or short backwardized contracts.

ColdFact

There are some papers on this subject of backward/contango that have been written mostly with respect to long only commodity indexes but in long/short portfolios too.

In general if you are long then backwardation is a "tail wind" and contango a "head wind" and vise versa for short position. Additionally if you are long a contract in contango in general the further out you are on the curve the less impact that contango will have, i.e. a lighter head wind.

Perhaps take a break from the keyboard and dig up some reading materials and you will see in general why this is the case. Energies are the classic example and one where the dynamics of the last 5 years have really changed and where using longer dated contracts for longs helps a lot.

Some keywords to use in a search is "optimal roll" for some basic information.

This makes sense, thanks so much!

The comments/advice are very much appreciated, going to read/research more

I've heard a lot of arguments on how "properly constructed" perp contracts should eliminate this issue. Yet I continue to see discrepancies like coldfact.

When TB is able to import raw contacts for testing, it will be a huge leap.

I always use raw contracts for backtesting and almost without exception, TF results degrade compared to using perp contracts.

Do you have a rough estimate in your experience, for how much the performance is inflated by using the perp contracts? I'd love to be able to adjust my expectations appropriately. Also, do you find that specific markets are more/less affected?

There was no firm figure for it, but it was big enough to cause concern.

It's a case by case issue. The idiosyncratic details of your strategy will also effect it.

Id love to be shown differently, but until then I would never base any conclusions on perp contracts. You gotta use raw contracts or some derivation scheme. Perhaps use perp signal days and then go to raw data to estimate PnL.

Call me a heretic, but I've seen this movie a few times before.

Bravochico,

Would it be possible for you to cobble together an example that shows how using raw contracts would give a different back-test result from using continuous contracts?

I am really interested to understand the mechanism, and whether the issue is just a case of using the "wrong" type of continuous contract, or whether the only solution is to use raw contracts.

Thanks

I'm cafe lounging in Europe at the moment.

When I get back, I'll dig up some examples but it will be a few weeks.

I'm not discounting the use of perp data, simply stating it's only a starting point and that I'd never put real money on perp data only. Never,ever.

Are you using volatility-based position sizing? If so, that could be something to think about as well. It's been my experience that deferred contracts often have significantly lower vol than front months, for various reasons.

depending on how your sizing algo works, less vol might mean more contracts.

may be worth researching

my guess would be that b-cat hit the nail on the head......

thanks for posting that data!

you could formulate and (using Blox) test a hypothesis:

• at the same level of risk, trading deferred contracts produces greater (gain-to-pain ratios) than trading front month contracts

Don't exert the effort to do this unless you think it's very important .

Bravochico, do you use TB to test individual contracts? I want to be able to do that as well but indicator creation get's iffy in some markets.