With the now constant intervention of the central banks in the markets, historical patterns of time-series behaviour would change perhaps significantly; any thoughts on how to go around this problem while modeling?

QuoteOriginally posted by: droneWith the now constant intervention of the central banks in the markets, historical patterns of time-series behaviour would change perhaps significantly; any thoughts on how to go around this problem while modeling?First of all, in the G10 space, interventions are quite rate (except for very few case in the Yen and swiss franc where the CB stepped in and sold its currency). When I run a "fair value" regression I usually takes a relatively long time series (say 5-7 years worth of data) to get around these unique situations. Also, given that a trading model will usually wait for a 2std< case you have sort of a safety net for cases of inverventions.Other than that, you have almost no way of getting around it, except if you can predict when the CB will step in (I know that in japan many look at the Yen TWI, not the USDJPY and the N225 reaction to the Yen strenght).

right now i am concerned with data sampled intra-day, say 15-30 minutes and the look-back period is few weeks or months at best. True that in G10, the intervention is rare, so probably not so much of an issue there. I would nevertheless want to get rid of the outliers and random shocks, as these seem to pose some problems in modeling, but don't seem to be able to 'clean' the data in a systematic way..

If you're looking at medium frequency sampling you will indeed suffer some issues. As the FX market is mainly driven by release of economic data (CPI, GDP, NFP, etc...) you can observ price shocks (gaps and jumps) surrounding the time of the release (usually US data are released at 8:30AM EST and 10:00 EST). The EUR an GBP data are released around 9,10,11AM CET. You, in my opinion cannot work around these gaps, as they are part of the FX market's dynamic.Just my 2cents...