Friday, 3 September 2010

What we've learnt in just the last two days . . .

Jerome has been working hard on the details of trade execution and has discovered one or two interesting things already. First of all, he placed around 40 actual trades, getting 38 trades more or less exactly on the expected fill. But in 2 cases, his fills were out by a cumulative 25 points, which is pretty poor. This revealed that his orders were taking a long time to reach the broker, and the price had moved by more than the broker's filter, so were rejected - by a "long time", we mean about 7 seconds.

This prompted a long series of tests of Jerome's computer equipment eventually involving the need to investigate the computer's registry. This is always a slightly nerve-wracking thing to do, but revealed, with the aid of a utility, that there were a number of registry problems. Fixing these and re-testing with the broker suggested that the time delay had been cut to less than 0.2 seconds. That might mean that we get filled more or less exactly in line with the market.

One of my "key variables" that still needs adequate testing is trading "slippage" - the difference between what we think we should get for a trade price and what we actually get. For modelling purposes I have assumed this is at least -25 points per day. But Jerome seems to think it should be lower than this, maybe even less than 10. A 15 pt gain on slippage would be excellent news. In about two week's time I should know the actual number to high accuracy.

Then Jerome discovered a method to load our trade size into pre-set deal tickets and to have these displayed all the time. Previously we had been pondering on the occasions when several trades are supposed to be done simulaneously and had assumed that they could take 30-45 seconds to actually fill say, three trades. But the pre-set display might cut this to as low as 5 secs to do 3 trades, so a major possible source of slippage might not apply.

Finally, my own research has suggested that we should only use the wide stop methodology rather than more complex methods that we have been looking at over the past few weeks. Each market still requires one more key variable to be set, but we have initial estimates which should be ok until the research is finished on this point.

Friday's economic data gave a good case study again in what we should be doing. We were the right way round in 2 markets as the news hit, wrong in 1 and neutral in the 4th. So good gains in a few seconds. But if we had been wrong, no stop strategy that we have looked at would have done better than the single, "big stop" idea.

We then discovered that Monday is actually a US holiday so trading will be lower than normal This could be a good thing, so maybe we will just trade as normal and see what happens.

No comments: