Exploring text to see what might happen in the stock market, part 2
Feb 21, 2017 8:28 AM
At the end of last year, I composed a blog entry focusing on picking stocks that would outperform the DOW jones industrial average over a quarter. The goal for the blog post was to pick five components of the Dow Jones Industrial Average that outperform the entire index. To make my selections, I used the JMP Text Explorer to evaluate earnings call transcripts over the past three years.
How did I do on my goal?
First let’s start with my overall goal: Did my five stock picks outperform the entire index during the quarter? For the start of the quarter, I picked Oct. 31 as that lined up with the release of earnings call transcripts well. The quarter ended Jan. 31. The five stocks I picked were CAT, GS, MCD, MRK, UNH. I “purchased” one share of each stock. The table below shows how each of my picks did during that time span.
Each stock I picked increased in value, and three of my picks had double-digit increases in that time frame. The summary of the performance is below.
My original investment of $570.61 increased to $671.64 in one quarter. This is a 14.45% return on that investment, which is a great return rate! However, anyone who has paid attention to the market over the past quarter knows that the Dow Jones index as a whole has done very well. Let’s take a look at the performance of the entire index during that same time period (see table below).
The index as a whole showed a large increase, going up 9.28% during that period. Had I used the same initial investment of $570.61 to purchase shares in the Dow Jones Industrial Average (assuming that was possible,) I would have had $623.56 on Jan. 31. That is still a fantastic return rate for a quarter, but not as good as the five picks made using Text Explorer in JMP. This is summarized in the graph below.
Great! I accomplished my main goal. By summarizing using Text Explorer, I was able to increase my return by more than 5%, resulting in an extra $48.08 on a $570.61 investment.
How accurate were the models?
I made 30 models to predict how the 30 components of the stock market would do based on the earnings call transcripts. A summary of my predicted close and the actual close are shown in the table and graph below. Some of the models did an excellent job of predicting the future stock prices, for example for CAT, PG and JNJ. Other models were not nearly as accurate, for example for GS, JPM and AXP. I found it interesting that the three least accurate models – for Goldman Sachs, JPMorgan Chase, and American Express – all were for the banking industry. Each model predicted lower than expected returns. Those three stocks showed the largest actual increase over the quarter as well. This increase in banking sector stocks could be tied to many different things. At the end of my original blog post, I mentioned a couple things I would have done differently to improve the models, including using more transcripts and other data not contained in the transcripts. Another big contributor to the model inaccuracy could be the election that happened during the quarter.
I accomplished my goal of outperforming the Dow Jones Industrial Average using earnings call transcripts. I was able to increase my return by more than 5%, which meant an extra $48.08 return on a $570.61 investment. I could have done better had I selected different stocks, but I think all investors can say that at some point. In some cases, the models made with Text Explorer were not very accurate, but they do add some information that helps predict future stock prices.