Volatility-neutral trading system

User Rating: / 0
Written by Forex Automaton   
Monday, 12 October 2009 12:47

After inspecting the simulated track of the best selected algorithmic traders (see EUR/USD, USD/JPY, GBP/USD, USD/CHF, USD/CAD, AUD/USD, it becomes clear that a volatility-neutral approach is needed. The optimized robots trade only during the peak of financial panic so there is a risk that if such a system is launched and the volatility returns to normal, no trades will be placed.

Trade decisions were made by comparing a predicted price movement (in relative units) with a threshold (discrimination against a fixed threshold). Initially when designing the system I did not anticipate that rare trading would become the optimal regime. I was aware that with a fixed trade idea discrimination threshold, volatility would become an incentive to trading, but considered that a normal thing. Meanwhile the situation when an optimized system trades only at the peak of the financial panic is certainly not normal. We need, ideally, a more or less constant stream of trading signals -- provided that conditions of non-randomness which makes the trades attractive are the same regardless of volatility.

  The fix I am currently implementing is to multiply the threshold by volatility. The ratio of a prediction to volatility which is effectively being compared to the threshold is closer to being volatility-neutral than the prediction itself. (In volatile markets, our prediction, naturally, also becomes volatile). The trivial changes of volatility with time scale are also taken out of consideration, so that a comparison of the results obtained with the same threshold among the different time scales (day, 4 hours, 1 hour and so on) becomes meaningful.  

The only argument against such a volatility normalization I can come up with is that conditions of non-randomness may not be the same regardless of volatility; the status quo is fine if more volatile environments happen to be more predictable. Which is not a mere conjecture. I remember reading a piece on statistical arbitrage (in stock markets) by Ed Thorp, where he mentioned that they had to exclude, for the purpose of model optimization, the market crash of 1987 and a certain chunk of data following this event as a unique money-making environment, too good to count on in future. Regarding the present crisis, a similar impression could be obtained from the comparison of crisis and pre-crisis forex correlation functions. However, I feel that this topic is not researched well enough for design decisions to be based on a hunch like this.

Bookmark with:

Deli.cio.us    Digg    reddit    Facebook    StumbleUpon    Newsvine
Last Updated ( Saturday, 16 January 2010 13:55 )