Quantopian's community platform is shutting down. Please read this post for more information and download your code.
Back to Community
Kakushadze Model for Overnight Returns

Hello all,

I've been working on some of the ideas proposed by the following paper: https://arxiv.org/pdf/1410.5513.pdf.

Essentially, the author proposes four factors to predict overnight returns. However, a little research and a quick backtest (see attached notebook) seems to show that these factors seem to be generally unpredictive... Even worse, due to the fact that any overnight positions must be exited the next morning, there is an incredibly high turnover, and the transaction costs will swallow up any potential profitability.

I don't see any way to salvage this idea, but I'd be very grateful for any comments or suggestions on where I should go from here. Should this idea be abandoned? Or is this a feasible idea for an algorithm, and if so, how might the factors be improved, and how might the algorithm be implemented?

1 response

you could run it on the time period proposed in the paper. If you get similar results, then perhaps you have competition that has taken this opportunity. If you get different results, perhaps there is an important flaw in your implementation.

The most important issue is the first one you list in the notebook, factors are not predictive. To deal with this, perhaps you can use machine learning on these and other factors to determine relevant factors for the overnight scenario. I'm a 25 year professional data scientist that is trying to do this.

The less important issue is that transaction costs are significant. My ideas include:

  • perhaps you don't sell securities that are in the first quartile of the next day's overnight hold period, except that this would only be known at the end of the day, not the beginning
  • perhaps the transaction costs as a percentage of the purchase price could be included in the decision whether to buy
  • perhaps the sell period could be studied to find a better moment than the open

Please tell me if this is still an area of interest for you.