Back to Community
Trading Strategy Testing

Hi All,

I am looking to get a better understanding of the steps one would undertake to perform a proper backtest or statistical validation/testing of the below theory.
What in your opinion would those steps be or what kind of proof would be sufficient to either validate or deny the claim.

Theory/Claim: Xover decompresses vs Main (the iTraxx CDS indexes) on a proportional basis when the market sells off.

As of now I have going around the idea of somehow defining what a market sell off is and then performing a hypothesis testing on the time series I have which is index spreads since 2005. Would that be a suitable approach or is there a more appropriate techniques?

Thank you

2 responses

Hey,

Unless I've misunderstood, since Xover is an HY CDS index, while iTraxx Main is IG, the relationship between the two indexes is not proportional (on a linear basis) during market stress. The spread between the two indexes tend to increase during these periods as the HY index bears more risk.
The main drawback that I see is that the two indexes are not predictive one another, they react at the same time. Trading the idea that during a sell off Xover will increase more than Main require to determine the level at which you consider that the market enters a sell off...
Another idea would consist to trade the idea that the spread between the two indexes will start to narrow once we've reach the peak of the stress. But once again it requires to determine the spread levels on which you enter and exit the trade, and pray that another stress will not appear during this period...

Hi Mathieu,

Thank you for the swift reply and input. I am absolutely with you on the notion that for this trade to work you need to assess when a market sell off has occurred and that is a challenge on its own. My plan for this is to test a few scenarios and determine which one really works.
Lets assume that has been done, how would you approach statistically proving that the decompression occurs and its significant and not random? What kind of stats technique would make sense give the daily time series data?

Thank you