Home
Welcome to the knockoff_index wiki!
This project aims to generate positive alpha by using an econometric tool –knockoff filter as in Barber and Candes (2015). The top 100 US stocks by market cap are considered as the candidates to construct active portfolios. The portfolios are based on analysis of the past 10 years performance and rebalanced annually. The performance of portfolios are measured by excess return (relative to effective funds rate) and are tracked over 1996-2020.
- This is a low-cost econometric-based approach to create a positive return in the stock market.
- This strategy is reproducible for non-US stock markets. If you have such data and are keen please get in touch.
- The investment strategy can be linked to certain investment factors or goals. For instance, the same approach could be used to consider firms with the highest ESG scores. The latter creates active portfolios that generate strategies that do not sacrifice return for green investments. If you have ESG data and are keen please get in touch.
The knockoff filter develops a regression model by controlling for false discoveries. This project uses four alternative US indexes – DJIA, S&P 500, NASDAQ, and Russell 1000 – to pick an optimal subset of qualified stocks. It is up to the practitioner to define the qualified stocks. For the moment the pool for qualified stocks is formed as 100 US stocks with the highest market capitalisation. The stocks picked by the knockoff filter are used to build optimised portfolios with minimum RMSE in forecasting the benchmarks (i.e. the four stock indexes).
See the codes and the accompanying read me file for a step-by-step guide to replicate the results.
Coming soon. The selected stocks and recommended weight on the portfolio will be released under folder "Portfolios".
The figures below backtest portfolios with a nominal value of 100 on the last trading day of 1995. All portfolios are assumed to be financed by effective federal funds. In Panel A four knockoff portfolios are exhibited against four US stock indexes. For instance the knockoff_DJ portfolio is formed for each testing period (1996-2020) by looking into the past 10 calendar years worth of trading, picking the top 100 stocks by market cap, and then further knocking off the stocks that do not contribute to explaining DJIA. Finally, the optimal weights for the knockoff_DJ portfolio is derived by minimising RMSE to predict DJIA in the past 10 years.
Panel A – portfolios picked by knockoff filter and optimised to match the underlying benchmark.
Panel B – reduced portfolios picked by knockoff filter optimised to match the underlying benchmark.
