We're seeking a prop shop which would allow us to trade $10m of their money, to obtain "live" results of our backtested daily Rachev ratio/zero investment strategy to raise a fund to trade it and to fund our development of a one-minute one. We ran the daily with the parameters of previous 365 days of returns, rebalancing every two weeks, equal portfolio weighting. We used a universe of 470 stocks to ensure sufficient return history and chose the top 47 and bottom 47 winners and losers. 365 days seemed to be an insufficient amount of returns (there were some backtest errors that occurred), so we used the previous 500 days instead. We also used exponential smoothing with a factor of .995. Unfortunately, we were not able to implement the noon trading time restriction because of a lack of intraday data over the full testing period. For the period September 2012 through September 2013, the portfolio lost 3% of value. For the period September 2013 through 2014, it lost roughly 1% of its value. This includes a good approximation of transaction costs (we have end-of-day bid-ask data.) For the period September 2014 through September 2015, the strategy gained 14.64%. All of these results are with no leverage, so with an Interactive Brokers portfolio margin account, we would multiply the results by 6.
Huh? Multiply by 6? OK, Houston, we have a problem. First of all, throw the return away. Convert it into units of risk so that your return will be denominated in risk units or sigma. That way the conversation about leverage goes away. Two, run a monte carlo against it where you have a fake strategy running with the exact same parameters in the exact same environment only with random entries. Run 1000 iterations and report the mean of the random return. Get the sigma of the fake portfolio and set your actual strategy's return to this sigma value and compare the two returns. The difference is your true alpha. THAT is what an investor wants from you.
aucociscokid, I’m not sure you are doing the math correctly. On a long/short portfolio that is dollar neutral, they don’t offsets for margin . The longs and shorts would each get shocked by at least 15%. In a portfolio margin account with $1mm, the most the OCC haircut will allow is $3.3mm long and $3.3mm short or 6.6:1 total. In actual trading at any clearing broker, it is very hard to really get 6.6:1 because of house risk rules. My best guess is you can aim for $2.5mm long and $2.5mm short and even that might cause risk issues. Bob
Thanks Maverick74 and esp. rmorse. Will recalculate according to Maverick74 and post. Thank, again. ALL comments solicited, welcomed, and appreciated. You guys are VERY knowledgeable and great! Thanks, again.
What is the value of backtesting and paper trading? Are those helpful for our own internal R&D process and denominating in risk units as Maverick74 suggests more useful for marketing purposes? How should be be doing our R&D? With backtesting or using some other methodologies? N.B. We're academics, used to that world. Is it that there is very little value in back testing because we all know what happened? We can build a system that works for the world we had, not the world we will have. Forward testing might have more value, but does not properly account for slippage, fear and greed. You never know how you will react to a flash crash until you trade though one.
Zari, our chief scientist, was working on the true alpha as Maverick74 suggested even before I responded to his post unbeknownst to me. Be happy to post it in about 10 days.
Wow, multiply by 6 in a year. That is really impressive! I am keen to see what would be the true alpha. Please let us know.... By the way, I am building my own automated trading system where I can invest my own money. I am back tested using IB historical data, then forward test on IB simulated trading, then I probes the IB price by buying 1 stock/future then sell it again quickly which I will use to estimate the slipage (it will cost the transaction fees). Finally I will trade a small number of stock/future then increase the numbers. I am not sure why you need 10 millions to test, you can always start small, then you can multiply by 6... I am still back testing and doing simulated trading, I will move to real money in early Jan 2016. Good luck with your R&D.