The title really should be “finding the least efficient markets – Part I”. But that’s such a broad subject I will just focus on one little part of that here.

Greenblatt’s book “You Can Be A Stock Market Genius” outlined some intuitive ways to find inefficient markets – spinoffs, M&A, bankruptcies…etc. But one of the most common forms of inefficiency is right there on the stock chart – the trend.

If you visualize the price chart of a perfectly efficient security, what would that look like? I imagine it would have sudden gaps up or down as new information comes out, followed by flat lines in times of no news. This is because in a perfectly efficient market, rational investors absorb and digest the same information instantaneously, and that should immediately be reflected in prices.

But often we observe security prices that trend. Prices go up for 3 days in a row, a week in a row…and so on. To me that is proof that the market is not perfectly efficient - there are delays in information dissemination, interpretation, and actions on the parts of investors. Whatever the causes, the trend presents good trading opportunities.

The trend is your friend. But how do you identify securities that are most likely to trend? We need ways to quantify “trendiness”. Here is one simple way to do it: count the number of times where prices move in the same direction (“sequence”), divide by the number of times when prices reverse (“reversals”). This is called the “Cowles-Jones ratio” (CJ ratio).

For example if you have price time series data that goes like this: 1, 2, 1, 2, 1, 2. That’s 5 reversals and not a single “sequence”. The CJ ratio would be 0. On the other hand, if your data is this: “21, 22, 23, 24, 19, 18” That’s 4 times where price moved in the same direction and 1 reversal. (22, 23, 24 all moved in the same direction, then a reversal on 19, and finally 18 moved in the same direction as the last number). In the latter case the CJ ratio would be 4 / 1 = 4. So if you go long the security whenever price first ticks upward, your chance of winning is 4 times that of losing.

Calculating this number for SP 500 components from 1/1/2013 to 9/30/2015, I find the average CJ ratio to be 97%. I expected a number close to 1 so this is reasonable. Below are the stocks that with CJ ratios that are 2 standard deviation above the mean – i.e the trendiest stocks since 1/1/2013.

So the “trendiest” stocks have CJ ratio around 1.1 – 1.2 range. A simple strategy would go something like this: go long whenever you see prices shift directions and go up; and short if prices reverse and go down. Your win percentage would be better than 50/50.

Now check out the common currency pairs. Total trading days are more than those for stocks because the stock market get various holidays off.

Note that even the least trendy FX pair is more likely to trend than the trendiest of stocks! That makes sense to me. The forex markets are full of non-economic players like central banks and commercials for whom profit maximization is not the top priority. Then you also have mom and pop participants. When my dad wants to buy some NZD he literally goes to the local banking branch and buy them! That surely creates lags and opportunities not seen in the stock market.

Now check out the common currency pairs. Total trading days are more than those for stocks because the stock market get various holidays off.

Note that even the least trendy FX pair is more likely to trend than the trendiest of stocks! That makes sense to me. The forex markets are full of non-economic players like central banks and commercials for whom profit maximization is not the top priority. Then you also have mom and pop participants. When my dad wants to buy some NZD he literally goes to the local banking branch and buy them! That surely creates lags and opportunities not seen in the stock market.

These numbers change depending on what time period you use. But in general I do find currencies to be trendier than stocks.

There are other ways to measure trendiness – perhaps one can quantify autocorrelations, or run backtests using simple moving average crossover rules and then rank the results. As I learn more ways to detect trends (and get more mathematically skilled) I will post my discoveries.

There are other ways to measure trendiness – perhaps one can quantify autocorrelations, or run backtests using simple moving average crossover rules and then rank the results. As I learn more ways to detect trends (and get more mathematically skilled) I will post my discoveries.

## No comments:

## Post a Comment