Above and below you can see some recent examples of stocks that went up a lot, and then either kept going up or dropped down again. Yes, that amazing analysis is given to you, free of charge.
Riffing on my previous ‘peak crash’ research, I thought I’d look into what happens to stocks that have a big bump up, either as a steady increase over a few days, or a more radical single-day increase. The results, at least from a shorting standpoint, were very pleasing.
First, the parameters:
• Period of 1/1/2000 through 12/31/2015.
• historical constituents of the Russell 3000 index (no survivorship bias).
• Historical price was over $10 and median 19-day trading volume was over 100,000 shares.
• A signal was generated when a stock averaged 3% increase per day over 5 days. This could be achieved either in a large single-day bump, or over the space of a few days. A stock that was completely flat for four days and then went up 15% on the fifth day would generate a signal, as would a stock that increased 3% each day for five days. To avoid overlap, I also required that the previous bar did not also have the same signal.
This gave me over 54,000 signals, which ought to be enough data points for most people. I recorded the forward 5, 10 and 20 trading day gains/losses for each signal.
By the way, I chose percentages rather than some multiple of average true range, because my underlying behavior theory is this: when people see a short term gain of 15% for a particular stock, whether it’s over a day or over a week, many retail investors would be tempted to lock in the profits and sell. Whereas few would say to themselves “hey my shares of XYZ are up 7 times their 15-day ATR over the past five days”…most people don’t think like that. Nothing magical about 15%; it’s an arbitrary amount that most people would consider significant.
Here are the overall gain/loss averages for the different forward-looking periods.
5 days: -1.041581293 % 10 days: -1.057296398 % 20 days: -1.070156631 %
Hey that looks promising! All periods show an average loss of 1%, which is meaningful over such a giant sample size.
As with the previous ‘peak crash’ research, I then examined overall market health to see if it had much of an impact on the returns. I recorded the same “S&P500 percentage under the highest close of the past year” value for each signal. I sorted my big list by that, and divided up the forward gains into vigintiles (20 quantiles).
Hmm, doesn’t seem to matter much how the market is doing. Except in the lowest depths of a bear market (to the right of the graph), in which case things get very unpredictable. For less extreme markets, there does seem to be a tendency toward better shorting returns as the market falls. But I don’t see this as a very strong tendency.
Let’s take a look at the size of the ‘bump’ then. Do the returns get better or worse for shorting as the 5-day average gets higher in the data?
One thing that caught my eye: the difference in returns between a 5 day holding period and a 20 day period aren’t all that different. Let’s generate yet another graph to help visualize that:
Above you can see the difference between the 5-day and 20-day average for each vigintile. The biggest is just over 0.6%, and most are much smaller. This means that holding (and thus increasing our risk) over a longer period doesn’t benefit us with a significantly bigger return.
The short-term market movement can have a big effect on an individual stock’s movements. If a signal is generated, and then the market keeps rising in the following days, then the stock is more likely to keep rising too (thus killing our short). And the opposite is true: if the stock goes up, we short it, and then the market has a few bad days, then our shorted stock is more likely to fall as well.
The market tends to be mean-reverting on a short term basis. Might we see a better result when we short a stock during a market upswing, knowing that the market will likely go back down in the following days?
To test this, I used an extremely simple filter. I divided up these signals into two batches: those that were generated when the S&P500 was above its moving 5-day average, versus when the S&P was equal to or less than its 5-day MA.
Shorting stocks when the S&P is up looks like a good idea, as the first chart looks solid. Shorting stocks when the S&P is below looks horrible! The longer-term holding period of 20 days does ok, but the short-term periods fall apart. Let’s look more closely at the 5-day period when comparing an up vs down overall market.
Obviously this isn’t a trading system, since there are no entrance or exit criteria. But it might give you some ideas for future short trades. It’s given me some ideas, that’s for sure.
Update 02/03/16: there is more to this story, and I’m going to do a second post on the subject soon.