Better Know Stochastics

I’m going to be hard on stochastics. If you love them, try not to take it personally. Of course, no indicator is a holy grail, but in this case I’m going to argue that you can often beat the stochastic oscillator just by eyeballing the bare chart!

In short: I don’t think fast/slow/full stochastics are useful indicators. I don’t suggest you use them to trade.

Those Alluring Stochastics

I tried to like stochastics. They are certainly charming in their simplicity. They draw nice clear oscillating lines, which look decisive and informative. I doubt you can find many new traders that couldn’t explain that 80 is overbought and 20 is oversold, for instance. Furthermore, they have sophisticated-sounding “%K” and “%D” lines, which always gave me the impression that fancy mathematics were behind the results.

Here’s an example chart, which I just picked at random:

One of the most basic signals they give is to buy when they rise through 20, or sell when they fall through 80. Another basic strategy is to consider buying/selling when the %K line crosses over/under the %D line. There are other more advanced ideas like spotting price-to-stochastic divergences, etc.

When they work, they seem to be magic. As such, I spent a long time staring at charts with stochastics, looking for the combination of factors that gave reliable signals.

Finally, I looked up how to compute them. Shortly after that, I banned them from my charts forever.

What it Really Tells You

The two things you should always ask about an indicator are:

What aspect of the price and volume action does this indicator highlight? What aspects are hidden by the indicator?

You should only use indicators that highlight information you actually want to see. So, what the heck is a stochastic pair of lines? We’ll describe the Fast Stochastic here.

The %K line is just the location of the closing price, relative to recent highs and lows. That’s it! So, let’s take the standard 14-period Fast Stochastic. If the current candle closes at the high of the last 14 candles, then the %K reading is 100. If it closes at the low of the last 14 candles, it is 0. If it’s smack in the middle, the reading will be 50. Simple, huh? The %D line is just the 3-period SMA of the %K line, so it’s just a slightly smoothed version of it.

To put it another way, the stochastic is just the percent position of the price in a donchian channel, plus a smoothed version of that position. I don’t know about you, but when I realized that, I felt more than a little misled by the complicated-sounding %K and %D nomenclature!

To get the classic buy signal (a cross above a 20 reading), the price just has to come off its recent lows a bit. Now we can ask: In what scenario would that actually be a good time to buy? Well, for the 14-period Stochastic, it would work best if the stock were oscillating in a channel with a cycle 14-candles wide or shorter. This explains why articles often point out that stochastics work best on choppy markets.

But, the accuracy of the signals is obviously greatly reduced if the cycle length of the price oscillation is larger than the period of the stochastic. For instance, what if the stock has been making a near-perfect sine wave every 18 candles? Now your stochastic is telling you sell because of a little dip relative to the high of the last 14 candles… whoops! Because it’s blind to the even higher high 18 candles ago, it told you to sell into noise just before price continued rising. We’ll see an example of this horizon effect below.

Do you have any reason to believe that your stock is in-synch with a 14-period measurement (or whatever number you used on your chart)? Neither do I…

A Couple Examples

Knowing this, it’s easier to see when signals are meaningless. Check out that IBM chart I gave above. See the sell signal between June 18th and June 25th, which didn’t work out very well? You should know right away that it’s not to be taken. Let’s say you presume this stock is in a channel (I’m not sure you should presume that, but let’s just go with it). With your naked eye it should be easy to see the top of the recent channel is around 108 (corresponding to those two peaks in late May). The 14-period stochastics are giving a premature sell signal because those peaks are more than 14 candles ago… That’s the horizon effect we mentioned earlier, where any cycle that might develop has a longer period than the stochastic you are using. This situation makes the indicator give false signals.

A signal that worked out well on this example is the buy signal from around June 11th. The only reason to put faith in this signal is that the lows correspond to a consolidation area in early May. The stochastics themselves are merely telling you that the price is rising from a 14-period low. Nothing you can’t see clearly in the candles themselves.

Are the Other Variations on Fast Stochastics Better?

Slow Stochastics are just a smoothed version of Fast Stochastics, so they inherit all the problems we’ve described, on top of the burden of additional lag behind the price. Full Stochastics are just parameterized around the amount of smoothing that takes place. See the
stockcharts.com article for full explanation.

Summary

So, there’s three main reasons I reject Stochastics:

• The basic signals only “work” when a stock is chopping in synch with the period of the stochastic you chose
• The stochastic isn’t telling you something you can’t see in the candles themselves (which is basically that it’s coming off the top or bottom of its recent range). In fact, I claim you can usually eyeball a stock’s recent cycle highs and lows, no matter how many candles ago they were. This means any off-the-cuff “stochastic” you estimate is likely more accurate than a fixed-cycle stochastic drawn on your chart.
• Even when stochastics apply and are working, it seems arbitrary to me to wait until the price is 20% off its lows or highs before I enter. If I think the stock is reversing in a channel, I should get in as soon as possible, shouldn’t I?

I’m sure some people who swear by stochastics might object that there are more sophisticated ways to read signals in them. Granted, but it must still be very sensitive to the period chosen. I don’t see a way around that.

Others may point out that all indicators are only showing you things that you could see in the price action alone. So, why should I give stochastics demerits? My answer is that I demand that my indicators show me something that is not obvious. Consider even the lowly moving average: I can’t tell what the exponentially smoothed price of the last 20 closes is just by eyeballing the chart. So, I use EMAs when I want to know that information. On the other hand, if I want to know where the price is relative to the recent high, I can do better than stochastics just by eyeballing the chart. This is because I can look back however many bars are necessary to get the right answer.

If you use stochastics, I’d love to hear about ways you deal with their shortcomings.

References

• StockCharts.com description of Stochastic Oscillator
• Investopedia definition of Donchian Channel
• Tips on Technicals Article comparing Stochastics and RSI.