There are days – like today – when many traders and investors may be wishing they ran a hedged portfolio!
Since our last entry our developers have delivered a core algorithm that runs over 25 full cointegration calculations per second – that is to say all unit root tests per pair, the cointegrating regression and the error correction mechanism. Better still, volume batches can include over 100,000 pair combinations for 36 month date ranges (we ran 481 pairs which produced 115,440 combinations).
If you have no reference point that will not mean much. But if you do it may pleasantly surprise.
The key to such capacity – especially for a batch of 100,000 pair combinations – is the use of a pre-screen filter. Instead of dumping all results, including the fails, into memory the pre-screen lets users define their searches by, for example, half-life, lags, mean reversion coefficients and proprietary normality filters. The tighter the criteria the larger the batches that can be achieved as the algorithm analyses and discards failing pairs thereby maximising memory limits.
10 year scans are also improved but more data means less speed – rather than 25 pairs per second a decade’s worth of data comes in a 9 pairs per second.
We are still on schedule and expect to contact pre-release beta testers who signed up via our home page during the third week of October.