Hi Paul,

Impressive result for your first algo!

How do you avoid look-ahead bias during the hand-picking of stocks?

Aidan,
I'm applying the same philosophy to stock picking as I would it I were just holding them in a standard portfolio; so a combination of financials, understanding what the business does, going mainly with solidly established businesses and the like. Since my intent with this part of the strategy is to buy and hold over the medium to long term, I'm not so concerned with stock performance on a day-to-day, week-to-week or even month-to-month basis. In other words, I am not sure if I have avoided look-ahead bias, I'm just investing in stocks where I trust the business, where they have had historic good performance and where I believe that they will continue to grow over the medium to long term.

The intent is for this to be part of my overall strategy, although there are definitely tweaks that could be made to improve the algo!

Hi Paul,
to compare apples to apples I ran the backtest as buy and hold, it seems the alpha is coming from stock selection

The issue I see with the backtest is that the stocks are hand-picked based on information we have today. We know these companies performed well since 2013 so you are picking stocks based on information you would not have at the start of the backtest, which is essentially looking into the future. I guess you could avoid this issue by using the fundamental data from before 2013 to pick the stocks.

Is there an inherent issue with using algo trading but hand-picking stocks? I ask because I am (very) new to algo trading, but have a little bit of experience with manual, hand-picked investing, almost exclusively as 'buy and hold'.

Is the overall philosophy of algo trading to be stock-agnostic and to simply use a combination of various signals and mathematical decisions to drive automatic trade decisions? If that is the case, is there a point where the hand-picking of stocks based on performance, fundamentals and finance comes into that, or should I aim more at identifying signals and strategies and not look at the more traditional indications on whether specific stocks are good investments?

In other words, should the algo do all the heavy lifting, or do I allow my own opinion on what stocks would be good investments to play into that?

There is no issue with hand-picking stocks, seemed to work ok for Warren Buffett! The issue is that you introduce bias in the backtest by using information that would not have been known or available during the period being analysed. This can lead to a lot of overconfidence in future performance.

An alternative approach might be to look at the fundamental data from before the backtest period and pick stocks based on simple rules (P/E < x, EPS > x etc.) for your portfolio.

Aidan, that's a very fair point. I have attached a backtest for the three years previous (Jan 2010-Dec 2012), and it doesn't get quite the same bang for the buck, although it does still outperform the benchmark. The alternative approach that you suggest is worth looking into as well. One of the reasons that I prefer to hand-pick stocks is that I prefer ethical/socially responsible investing; I wonder if there is a way to lift data on socially responsible businesses and incorporate that into an algorithm - That could be worth another post...

Thanks for sharing Paul!

If anyone else wants to test rebalancing strategies with custom-stock portfolios, take a look at our new tool AlgoBuilder: https://www.quantopian.com/algobuilder

It allows you to choose a pre-selected basket of stocks or create your own portfolio from over 8,000 stocks and ETFs in our database. Assign customizable weights and see how the algo performed over the last 2 years. Backtest the strategy without writing a line of code!

What you could do which would be interesting to see. You said you have been a buy hold investor for quite some time. I don't know if Quantopian software will let you do this but you could look at your trade history in your personal account as far as when you made certain buy decisions at what time etc. and implement those stocks that you made buy decisions for in the backtests during the time period in which you bought the equity in your real account. Then no bias because you can justify putting the equity in there because either way thats the equity you decided to buy at the time.