Beware curve-fitting…

We ran a few scans using aggressive filtering criteria in order to produce baskets of raw pairs we could back test. The focus was on achieving high levels of normality pair-by-pair.

No economic sense checking was performed on the results – we just wanted to test the power of ArbMaker’s statistical filters and run what they came up with against a uniform, untweaked and non-customized set of default back test criteria.

The results (which are total, not annualised returns – a simple 5 pair sample leaves significant “dead money” date blanks between trades):

Such an in-sample is, of course, too small to draw earth-shattering conclusions from but our main goals here are to:

1. Show how ArbMaker’s stat filters blow away the at least 80% chaff users would otherwise be wrestling with and harvest a core list of pairs for analysis.
2. Demo how even the default back test criteria, linked to multi chart factors, churn out profitable entry and exit points.
3. Reveal some of the portfolio-level features we plan to roll out in Q1 next year that will make this kind of testing easy to do.
4. Suggest that, run right, stat arb builds-in very stable hedges and reduces draw down risk.

The pair combinations used are listed above and all the raw data shown can be pulled from ArbMaker for replication purposes.