Execution errors evaluation for low frequency systems

[cordura21]

A situation occurred to and I'd be very grateful to have your opinions.



We have a low frequency trend following system, which is unlevered. A person checks the signal on an Excel spreadsheet with Bloomberg's API and enters a trade. This runs once a week, so it's pretty basic and we didn't automate it.



There was a change in the API parameters so the data to calculate entry points was wrong and we enter late. There is no way of doing forensics, no logs on our side or data providers'.



It ended up that the trade was a profitable one, so we had an opportunity cost of 15 bps on the AUM given the subsequent price path.



Now I know it's a failure on the design of the procedure,I guess both on the data gathering procedure (automation) and on the data checks (human). Trader didn't even know there were preferences that could change.



So here are mi thoughts on how to measure it to define the seriousness of the situation, from more to less serious:



- don't measure it as money, it was a design mistake

And then also:

- measure it as a considerable nominal amount of money, thus very serious

- measure it as a big hit on the assets, thus very serious

- measure it as a milder hit on the trading (since it rotates capital more than once a year)

- weight against the changes of having a good trade, if you consider that trend following has a 40%.



What are your thoughts on this? Do you have a protocol or procedures in this kind of situations?



Cheers, Cord

[apine]

What is the motivation to measure the seriousness? Doesn't it just make sense to fix the problem by putting in some automation and/or sanity check figures?



I think that the fact that it is unlevered and low frequency is irrelevant (although I totally understand why you mention it). This is a business risk not really a market risk. If the trade is not worth doing, then don't do it. If it is worth doing, do it right and put the appropriate controls in place for it.

[cordura21]

We already put sanity checks in the procedure. It is not worth to automate much, specially since our automated platform is made for futures and options and not dividend paying equities (dividend adjustments was the culprit).



The reason to measure the error is to differentiate between strategy return, slippage and our platform execution errors. This last part is to assess if our procedures can be improved. And specially to check if there was a manual modification of the parameters. Although I know that you need to have a process good enough to avoid this kind of event, having a foolproof process is costly and some strats are ok tbe manual. And then you need to assess the people.



Edit: i meant that we already put sanity checks