Unnaturally Long Attention Span

AvatarA blog about Grad School at Stanford, Working in a Silicon Valley Internet Company, and Statistical Machine Learning. mike AT ai.stanford.edu

On Stock Trading

I've made quite a bit of money this past week trading stocks in my Roth IRA. Here's a few things that I learned, which I hope will be of use to you as well.

1. The reward function of playing the stock market game should NOT be the stock price itself. The reward function is actually SELLING stocks at a high margin over what you paid for them. So, if you are training a reinforcement learner (such as your brain or a Markov Decision Process) , no reward should be applied when the stock price increases, unless a sale is to be made.

Taking this into account, only two modes of short-term stock trading make sense to me:

1. Betting on a Bull Market
In this mode, you begin the trading near the beginning of a big event, such as the winter shopping season or the end of the fiscal year for a particular industry. You choose a basket of stocks that reflects how you predict the industry will be changed at the end of this quarter. Take all of the companies that you think will be affected positively, subtract out those which everyone else thinks will be positive (since those expectations are already reflected in the price) and buy those stocks. So for example, if you think AMD will beat out INTEL in the next batch of processor releases and it is counter to what the public believes, then buy it! Hold on to the stock until the event passes, and then sell your stock at the end of this 1-2 month period. Your gain/loss will reflect exactly how good your prediction was.

2. Interday/Week trading
The most important thing to remember is that each trade costs you $7 (with Scottrade, which is what I use), so you must try to make trades that at least cover you for the $7. This means that the trades necessarily have to be in larger volumes on stocks that have higher volitility. This translates into more risk. However, there are techniques to balance the risk by choosing a set of stocks that balance each other. Unlike the first mode, this type of trading can be done even when the market is flat, since you are just betting on the small changes that happen day-to-day. In addition to choosing stocks that move a lot from day-to-day, you want to choose stocks that have a sufficient information streams. What I mean, is that even though those penny stocks might have huge amounts of movement, if there is not much news related to that company, you don't really have enough information to make an informed trade. So, you really want to choose companies that have a mid-size stock, information-rich news stream, and those that are not likely to go out of business the next day. For this type of trading, you want to hold onto the stock for at most just a few days.