Force Rolling Vs. Close for Dynamic system

[oldlore]

The idea of closing a position about to expire instead of rolling it and allowing the dynamic system to decide whether to re-open it, do nothing, or open a new position has come up before. I don't think I've seen it discussed much. I see in his report repo that Rob is doing this in his own system. I've been playing with it myself and had some observations and questions.

-I've cloned my system on another computer. Let's say GOLD_micro is about to expire. I'll close it on the clone and then run the system to see what it would do. Now in reality the live system would have 1 new day of data compared to the clone. But I've noticed that the clone most often re-opens the same position.

-I've noticed that if I close multiple positions on the same day, the system is much more likely to choose a somewhat different configuration than re-open the exact same positions.

-If the goal is to allow the system to choose positions that get it closer to the optimal portfolio, then does it make sense to close positions in "batches" when expiries and liquidity allow? When force rolling it didn't really seem to matter when you rolled one instrument vs. another.

[robcarver17]

It makes sense that if we make just one change to a position vector, and that pushes it outside the buffer, that the optimal DO response is to revert to the original vector.

I guess we can think of the 'close' option where the system re-opens the position the next day as being a bit like 'force_leg' (generate two seperate trades to close the existing position, and then open a new one), except that there is likely to be a day between the two trades. Clearly that isn't optimal, and I think you're right that the idea of 'chunking' close orders together (eg we're now getting close to the U expiry in a lot of markets) makes a lot of sense as it's more likely to cause the system to do a general reappraisal of optimality rather than just reverting to the prior position vector. As a general rule I tend to do this by accident rather than policy, since I usually review my roll status on monday mornings.

But in general I think it still probably makes more sense to use 'force' which generates a spread trade rather than 'close', when the market has cheap, highly liquid spreads.