Hi y'all!
Talked about this a few times here (usually after a few beers, not the case now
. I'm developing an application for backtesting option strategies and it's already in usable form.
Initially I've put it free and open source on GitHub but then I changed my mind and made the repository private: I fail to see the benefit of offering further chances to my competition, and the competition is brutal. I posted here a shapshot of a LinkedIn job for a quant developer which already had 1048 applicants at 12 hours after being posted.
Anyways, I've ran the first backtest and already I'm surprised by the results One feature of the application is to generate "theoretically pure" market data and run the backtest on it - in this case the results are beautifully aligned with theory. It may seem artificial but it's actually necessary in order to rule out software bugs: in the first runs the results WEREN'T confirming the theory, so I knew the problems are in the software and not the data, found the problems, fixed them and got perfect theoretical results.
Armed with the certainty that the software is working correctly, I backtested the same strategy on actual market data... and disaster!
Well, there still was a problem with the data: some of it it's effectively garbage and it's exactly the garbage that fucks up the tests: good data is usually arbitrage-free so it's ignored in the backtest, but crap data presents itself as an arbitrage opportunity - only it's not real but a fake. So I had to curate the data and managed to do it to some degree but I'm still not 100% sure it's perfect. That's because due to the huge amount of data I do the curation automatically, which means the result is only as good as the algorithm. Eventually I plan to do some painstaking manual curation, but for now I'm good with "probably mostly reliable data".
So even with this probably quality data I had a shock: THE MARKET DOESN'T OBEY THEORY!
What happened is that when trading arbitrary stocks, my strategy got destroyed by jumps in the stock price - legitimate jumps, not garbage, I cross-checked a few cases with market data from Yahoo Finance.
So now I gotta incorporate this knowledge into the strategy and re-do the test. Will tell you how it goes
Talked about this a few times here (usually after a few beers, not the case now
. I'm developing an application for backtesting option strategies and it's already in usable form.Initially I've put it free and open source on GitHub but then I changed my mind and made the repository private: I fail to see the benefit of offering further chances to my competition, and the competition is brutal. I posted here a shapshot of a LinkedIn job for a quant developer which already had 1048 applicants at 12 hours after being posted.
Anyways, I've ran the first backtest and already I'm surprised by the results
One feature of the application is to generate "theoretically pure" market data and run the backtest on it - in this case the results are beautifully aligned with theory. It may seem artificial but it's actually necessary in order to rule out software bugs: in the first runs the results WEREN'T confirming the theory, so I knew the problems are in the software and not the data, found the problems, fixed them and got perfect theoretical results.Armed with the certainty that the software is working correctly, I backtested the same strategy on actual market data... and disaster!
Well, there still was a problem with the data: some of it it's effectively garbage and it's exactly the garbage that fucks up the tests: good data is usually arbitrage-free so it's ignored in the backtest, but crap data presents itself as an arbitrage opportunity - only it's not real but a fake. So I had to curate the data and managed to do it to some degree but I'm still not 100% sure it's perfect. That's because due to the huge amount of data I do the curation automatically, which means the result is only as good as the algorithm. Eventually I plan to do some painstaking manual curation, but for now I'm good with "probably mostly reliable data".
So even with this probably quality data I had a shock: THE MARKET DOESN'T OBEY THEORY!
What happened is that when trading arbitrary stocks, my strategy got destroyed by jumps in the stock price - legitimate jumps, not garbage, I cross-checked a few cases with market data from Yahoo Finance.
So now I gotta incorporate this knowledge into the strategy and re-do the test. Will tell you how it goes
