One of the problems I have faced in the past is the appearance and disappearance of positions in a theoretical portfolio resulting from a back test used to generate real signals for real trading.
Thus having to face awkward decisions: do I realign my real portfolio with the changed back test? If I do not do this how do I track my real time positions which have disappeared from the back test portfolio....and so on.
For many years I have used CSI's futures data and the appearance and disappearance was usually due to retrospectively changed historic data.
Which is all well and good for data providers who need to correct mistakes or inaccuracies in the historical database but a fat lot of use to a trader who has already acted on such "false" data.
My colleagues and I are working on a python/mySQL combination which will append only NEW from the data provider/providers of our choice. Clearly before appending you may well wish to ensure there are no gross errors by using a simple algorithmic checking routine. Therefore, once a day's data has been accepted/approved, it will never change - at least in terms of the original unadjusted OHLCV. Naturally, the adjusted transform of such data will change with respect to splits etc for stocks or appropriate back adjusting for concatenated futures time series.
Do you guys adopt a similar methodology for the data used by Quantopians in their back tests?
