Hi,
in the current algo trading landscape the use of synthetic market data is becoming more and more common. Strategies can be made more robust by validating against a distribution of price curves rather than using a single occurence (the actual asset price curve).
Wondering whether there is a way within Zorro to generate it and use it in an efficient way...?

For instance, if I want to generate 100 synthetic price curves from a given asset (using TimeGAN / Tail-GAN / Sig-CWGAN techniques) do I need to generate 100 different t6 files (times each year of data) via custom script and call them individually, or is there a better way? Is there a native/embedded way to do this with Zorro?

If not, it would be great as a future feature to allow easy synthetic data generation and its use for testing strategies.


Thanks

If the data is anyway generated, you need not store it in t6 files. Just generate it on the fly in the initial run, and fill the price arrays with it.

That said, I know no way to generate synthetic data that could really improve backtests, except maybe in very special cases.

Interesting, thanks, didn't know I could fill the Price arrays from script.

Regarding synthetic data generation and backtest, I sugges this read (google translate): https://www-x--trader-net.translate...tl=en&_x_tr_hl=en&_x_tr_pto=wapp

Or the original website (in Spanish): https://www.x-trader.net/como-saber-si-una-estrategia-de-trading-esta-sobreoptimizada

What do you think?


Thanks

Thank you for the link. The problem is that trading strateges usually do not exploit statistical properties, but market inefficiencies. Autocorrelation is a market inefficiency, but volatility and return distribution are not. So this synthetic data will not work in the general case, only for strategies that happen to exploit just the statistical property that the generator replicates.