Saturday, October 13, 2007

Data mining

Today I got that idea to question the kind of instruments I trade. Is it worth to trade one contract or would my time be better spent trading another. Do I have any favorites? I had the idea to look at my performance by grouping the trades made by contract. I took the last year as I thought that to be a broad enough but still recent basis from which I might be able to draw conclusions, if there were any to be drawn. Within the last year I made 1689 trades, so the statistical sample is not too small I think.

I downloaded the last 12 monthly statements in Excel format from IB and extracted the trade information contained in the report.

I grouped them by contract ignoring the frontmonth traded and got the following list, which is already summarized and sorted according to P/L made per contract traded.

P/L MTM per contract = Profit/Loss Mark to Market per contract excluding commission

As you can see I traded quite a number of different contracts during the last year. And I'm sure I should concentrate on just a few of them. I already did that actually as I'm no longer looking at most of the contracts listed there. Still according to that list it might be worth changing the contracts I currently trade, as I do worse in some than I thought while I ignored contracts, where I seem to be not so bad at all.

Due to the diffent tick value all these contracts have, the P/L per contract will not give you the whole picture. You need to know, how many ticks on average (including all the losers) you make. So I calculated the ticks made on average. Sure it's just an average and it's a rounded result, but together with the Trades column it is able to tell you quite a bit. If the number of trades in a contract is very small, the result needs to be ignored. It's just a number of trades and the next 10 trades might paint a totally different picture. But once you have made 30 or more trades in one contract you can start to look at the results and expect the results to converge on a statistically significant result, which tells you how much Profit or Loss you can expect to make on average, if you trade a certain contract.

The most important information I got from this statistic is not what I should trade, but what I should not trade: I should avoid trading the FTSE and the EOE (Amsterdam index). And I sure have problems trading CL. On these three contracts I actually lose money on average with every trade I make. I also lost big trading ZG or NG. ZG is a kind of exception as in April this year I let one trade go out of hand huge, by adding to a loser and holding in the end 4 or 5 contracts, when Gold continued to go against me, stopping me out at a real big loss. Exclude this trade and ZG ends in the winner column


making me on average 21 ticks per trade.

It also seems to be worth trading COIL instead of CL. I know COIL moves slightly different from CL, but I never suspected CL to end in the red, while COIL was positive (the number of trades differ significantly which might explain the results of course). Well I will heed the information received and switch back to trading COIL for trading oil. GOIL looks interesting as well. Just that this contract trades with a 25$ tick value, which is a bit high for my liking, as you can't expect a contract to immediatly go your way. I also like the Loonie, whose results a screwed a bit by one month, where I seem to have made 18 1 or max 2 tick 6C trades. Exclude that month and you get:


I will add the Australian Dollar to the basket, as I remember saying for some time when I really screwed up: Hey Chris let's trade the AUD or CAD, it's a safe play to trade and you will be able to get your losses back.

What really surprised me was, that all in all I'm not so bad trading the ER2 than I thought. It's positive, paying mye some money on average per trade, so it's worth trading, something the HSI is not.

Tomorrow I will adjust my Ensign trade sheets, so I'm ready on monday to trade what I'm statistically good at, while avoiding trades in contracts I lose on average.