Hi all, this is my idea, which I have written my master thesis about (which is pretty much print-ready, but not final). I am currently in the process of implementing it and plan to use it primarily on the forex market, sometime later maybe also for stocks. The thesis talks about value investing because this was agreed upon with my teachers. Let me know what you think. This was started more than three years ago when my interest in automated trading began. I could think about working on this in a team, though not open sourcing it. See: http://invesdwin.de/public/Thesis.pdf Feedback might help me to further improve the text. I am open for discussion. Best regards, Edwin
Start with understanding the difference between Trading and Investing - "Value Investing" normally has long cycles - more than trading, which normally cares little about the fundamentals as much as the change in it (news). If Iinvesti n value, I want to wait until that value "matures". That is swing, long term. This "Trading" can be done with Excel etc. - semi automated.
> The thesis talks about value investing because this was agreed upon with my teachers. I wonder how someone can go about building an automated trading system using value investing strategies. To me the biggest hurdle is data quality because fundamental data are often inaccurate and have to be manually normalized/adjusted after going through 10Q/10K of each company. Most data from bloomberg/capital IQ/factset require normalizing. Add in different fiscal year-ends for companies and the whole exercise is practically impossible to automate. Any thoughts? Yana
So you think a backtest of an automated value investing won't be interesting for you to look at? Though I too think value investing only works in a semi-automated fashion, since the subjective criteria play a large role in the strategy. But what if it shows itself to also work in a quantitative only fashion? Do you think this has already been tried using the approach from Town? Anyway, I don't plan to implement the value investing strategy, I rather want to use the automated strategy development process to develop normal strategies like scalpers, trend followers, etc. I think it could reduce time effort by automating a large part of the trial&error which is normally done in manual labor. Also from my perspective I would not say that this is 1+ decade behind the market, since I have not seen a platform or a book that uses this approach for modular strategy rules to reuse code and generate combinations from variability models. If you have a reference I could look at, I would be glad for a link or a name of a paper/book.
In my experience while doing value investing manually for a few years now, you cannot use value investing without a trading approach. Even buy&hold which is often encouraged in the value investing area is a form of trading, though a risky one. You can improve that by using a breakout entry that waits for the stock to rebound, so that the entry does not catch a falling knife just because the company is already undervalued. In fact in page 9 I differentiate between a portfolio management strategy (most parts of value investing) and a trading strategy (which in my opinion can also work long term, since it only defines buying and selling of an instrument).
Yes, data quality is quite an issue. The sources I use are MSN Money and Yahoo Finance. They have their own quirks. Another site that uses their data is stock2own.com which I reference in the thesis. My experience is that for the commonly traded companies the data there is of high quality and good enough for automated trading. Though for unpopular stocks, the data sometimes makes no sense at all or has some errors in it. Since value investing as specified by Benjamin Graham mostly focueses on the larger companies, a possible filter could be to only allow automated trading of stocks that are listed in indices. Thus the data quality from these sources should suffice for a backtest. On the other hand value investors might be interested also in other companies or even the larger companies could have bad data. This is a risk that should be monitored, which is easier when doing this in a semi-automated fashion.
What you are describing in your "thesis" is basically a primitive form of evolutionary/genetic programming/strategy design.