As a sanity check, could you run your backtest on at least some actual data? Your brownian generated data may not be picking up the nuances of real data(?) such as price shocks, correlations among stocks and a multitude of other things.
Yes, sudden price shocks are not specially considered in this system; it rather works in "normal" market situtations,
meaning: the generated GBM data have the control property of a certain volatility, ie. one can compute
the volatility of the generated data and it will give nearly the same volatility that was given as an input parameter to the system, for example 30% in the tests.
But: the system itself does not make any use of that information; it is used only in the data generation subroutine.
And, I think the opposite is true like Sinatra said "if I can make it here, I can make it everywhere",
meaning GBM covers more outcomes/possibilities than real data of only some years.
But if someone provides me 5 sets of annual data of 67 options then I can test that too, but I think it is nearly impossible to find that much real data.
Regarding correlation: in the test, 67 tickers (underlyings) were used, and each was independent of each other, meaning correlation was not taken into consideration.
Maybe it should have been, but this would complicate everything.
The offer is just the result of a research using simulated but realistic data.
It does not take into consideration rare black swan events like market crash etc., except one thing: if there is no change in the last price then it thinks the price was halted at the exchange...
