I suspect the high price-to-book results could be from the tech sector which tend to have higher ratios due to the fact they have less book value to begin with, mostly computers, servers and facilities as opposed to manufacturing companies. Maybe limit it to a specific sector and see if that changes the results.

Thanks Jeff, that's a great idea.

I tried it again, filtering out technology, but it just made the discrepancy larger. It also doesn't seem to capture all tech stocks because amazon is considered consumer cyclical.

I think you can use context.account.leverage == 0 to prevent your backtest from leveraging.

I think you can use context.account.leverage == 0 to prevent your backtest from leveraging.

context.account.leverage tells you what your current leverage is; it doesn't enforce a leverage limit. You need to do that in your code.

And here's an example of how to control leverage (it's for the contest, but can be tweaked for your use case): https://www.quantopian.com/posts/quantopian-open-example-algorithm-to-control-leverage

I feel dumb for not knowing about account.leverage. Thanks! It looks like leverage does not explain the issues I am facing.

I ran the backtest with earnings yield, and ev/ebitda. These produce the proper effect in that the "value" end outperforms the "glamour" end, however, the "glamour" end does not underperform the SP500!

This is very confusing.

The reason value "works" is that when it's wrong, you get absolutely crushed. Note that your worst drawdown is actually worse than that of the S&P 500's. Yes, value can work "over time" assuming you have enough capital and enough screens and enough discipline to actually see it through. And at the end of the day, you're lucky to break Sharpe 1. Outperforming the S&p 500 in a backtest isn't saying much.

Don't feel stupid Johnny. I am having the same problem with my fundamental-based back tests, and context.account.leverage will not fix it. Attached is the backtest for my algorithm based on ev/ebitda. The algo is designed to buy and hold 5 securities with the lowest ev/ebitda but, as you can see, it completely breaks down over time. The back-test was run for a 10 year period (May 2005 to May 2015); by the end of the back test, on May 2015, the portfolio was negative $6.8 million in cash, and held over 80 securities.

Thanks for your reply John,

Im at work so I can't troubleshoot your code but I think part of the problem is that you are rebalancing daily. Value stocks need time for its market price to catch up to its true value. Also your market cap cut off is a bit too low. Lots of value stocks are going to be small cap illiquid and loss of capital is due to slippage.

Yes, you need to add code which looks in your portfolio positions for stocks which are not part of your universe (anymore) and close them out yourself. (if that is the problem I think you are coming across).

Simon, are you referring to Johns code or mine? I think we stole code from the same template lol. There is a line that is supposed to exit all positions that are not in fundamental_df. Its under the comment exit all positions before starting new ones.

Hi John,

Here is the same code that you posted, the only exception being that it now runs re-balance every year only. This reduces slippage and commission, but also gives time for the value shares to be recognized by the market and rise again. Not shabby eh.

Here's the same algorithm, but not magically excluding three sectors, and rebalancing mid-year (and plotting that it's exceeding leverage) There's a lot of ex post alpha in these fundamental ratios, but it worries me how sensitive they are to the precise rebalancing time.

Simon, John excluded 3 sectors probably because of the book Quantitative Value by Gray and Magic Formula by Greenblatt.

EB/EBITDA works well to compare between different companies because it is capital structure neutral. It takes into account debt levels. Utilities, finances and REITs report earnings in a more complex manner so they are excluded for that reason; their ratio are not comparable to those outside of their respective sectors.

I think I may have found the answer to my original question about why my low price to book ratio sort was underperforming:

Here is an article by O'Shaughnessy
http://jimoshaughnessy.tumblr.com/post/103140701394/price-to-book-value-ratios-a-long-term-winner

Here is a summary of the article, but has a neat graph suggesting concentration of risk to a small portfolio would have caused underperformance for the periods 2006-2013
http://investorfieldguide.com/2015210the-very-cheapest-stocks-pricebook-do-very-badly/?redirect=true