Elegant algo Andrew!

I was curious why your algorithm had flat returns until 2011 and it turns out that XIV began trading on Dec 10, 2010 and TMF started trading on April 29, 2009. I adjusted the backtest to run from Dec 10, 2010 when both of the stocks were on the market to reflect more accurate metrics. Pretty darn close to what you had!

If you're interested in iterating it further, take a stab at improving the drawdown. 30% is steep, would you be able to handle those downsides? Perhaps try this calculation: if your returns for the last X days were negative, close the position and when the returns are positive for Y days, re-enter the market.

It might be helpful from an intuition standpoint to compute and plot your effective leverage and market exposure, I'm not that familiar with XIV and TMF, but it looks like they are levered ETFs. You can use the record() function to plot custom variables like leverage or exposure over time.

Finally, if you're looking to live trade, you will need to check the margin on your brokerage account because this algo occasionally dips into negative cash (see attached backtest). This is due to the rounding of order_target_percent as it tries to adjust to your desired position. If you want to have a cash buffer (say 5%), then you can do something like this:

context.weights = 0.95/len(context.sids)  
order_target_percent(sid, context.weights)  

Thanks for sharing!

Very helpful response Alisa! Thanks!

I had not looked into order_target_percent in depth. I tried switching to floor rounding my orders. This reduced my negative cash, but did not eliminate it. I am guessing that by the time my order is executed, the price has slipped. A limit order at my chosen price would often not be executed. Using a cash buffer seems like a simple and effective solution.

I'll work on getting appropriate returns per risk in my next algo. Maybe a rudimentary pair trade using some TaLib indicators such as RSI.