Wednesday, November 01, 2006

What Happens After A Bearish Engulfing Pattern?

I notice that Brian Shannon, in his fine video review of the market, observed that Wednesday's action constituted a bearish engulfing pattern. Now I'm not exactly an expert on candlestick patterns, but the basic notion of a bearish engulfing pattern makes sense to me. We rise above the previous day's high, decline below the prior day's close, and close near the bottom of the day's range. On the face of it, that should be bearish, as we rejected higher prices. Brian notes that he doesn't put much stock in the pattern, preferring to focus on the longer time frame. In the end, as Victor Niederhoffer emphasizes, we must always do our counting and verify any conclusions with hard data.

I went back to 1996 in the S&P 500 Index (SPY; N = 2694 trading days) and found 45 occasions (like Wednesday) that met the following criteria: 1) today rose above yesterday's high; 2) today declined below yesterday's low; 3) today was the largest range of the past five trading sessions; and 4) today closed in the bottom 10% of its day's trading range.

Interestingly, the next trading day in SPY sports an average gain of .10% (31 up, 14 down), notably stronger than the average daily change of .04% (1405 up, 1289 down). Since 2004, we've only had 12 such bearish engulfing days, but--incredibly--11 of them have closed higher the following day for an average gain of .32%.

Clearly there is no next-day bearish edge to outside days that finish near their daily lows. Indeed, there has been a nice bullish tendency following such one-day declines. Once again, we find that anecdotal reasoning has its limits when it comes to market intelligence.


The Market Speculator said...

Great analysis, especially for somebody like me who trades off engulfing patterns. I would like to make a few points:

1-It's generally taught that traders use this pattern for spotting reversals in a trend. If the pattern occurs in congestion, or in the same direction as the engulfind pattern, it's useless. I'm not sure where the pattern occured in your analysis.

2-It is also generally taught that you wait for confirmation. Those 11 days would not have confirmed the pattern. Also, they may have confirmed in the coming days. When confirmation does occur, they can be very powerful.

3-Finally, the are supposed to work well when printing at support and resistance levels.

It would be interesting to see those 11 positive days in the context of these 3 points.

Brett Steenbarger, Ph.D. said...

Hi Market Spec,

Thanks for elaborating rules regarding the pattern; thanks also for your fine blog.

It would be interesting to test the various aspects of the rules to see if they do, indeed, help with the anticipation of future returns. One problem is that, if you get too many rules, it decreases your sample size and results become unstable.

Your post shows how difficult it is to test chart formations, as so much of their validity is dependent upon the context in which they occur.

In general, I found the pattern to give much better returns in bull markets than bear ones, with much of the gains coming during this recent bull swing.


John Wheatcroft said...

That's one of the problems with candlestick analysis - it must be taken in context. Now it is possible that your results are more closely related to another phenomenon that being the new highs/new lows that you have been speaking about - a bearish engulfing naturally results in an increase in new lows and a decrease in new highs so one would expect the S&P to react positively given that environment.

NO DooDahs said...

Additional points -

The candlestick patterns were meant to be used for, and were designed based on, the behavior of individual issues. Applying them to an index is not apples to apples. A better test would necessarily involve a very large database of stocks which were evaluated for a statistical pattern.

After a bearish engulfing pattern, basically a two-stick pattern, the typical candle-trader isn't interested in just the next candle's behavior, but in what the pattern signifies for a trend. So a proper backtesting would look further than just the next day.

I'm curious as to the intraday behavior you spotted in those 45 occurrences. Did they close lower than they opened, regardless of where they closed relative to the previous day's close? Or did they tend to charge up during the day? My limited work has led me to believe the indices are far more mean-reverting than individual issues are.

Where the index or stock closed in relation to the previous close is immaterial, because we have to observe the previous close to see the pattern. Our first actionable instance is the open on morning after the pattern is drawn. We can't both observe the pattern at closing and still carry the overnight risk -

Brett Steenbarger, Ph.D. said...

Hi John,

You are absolutely right...and the context may be quite different if the pattern is occurring after a previous decline vs. a previous rise. Your point about the new lows is also well taken, and one would expect the pattern to have different outcomes if it leads to expansion of new lows vs. a divergence. Thanks for the excellent points--


Brett Steenbarger, Ph.D. said...

Hey Bill,

Thanks for the excellent points regarding candles. I very much like your idea of analyzing those kinds of patterns on large groups of stocks. The indices, indeed, are more mean-reverting. I also agree that looking at the next day (and multiple day) patterns from open to close rather than close to close would be instructive for daytraders. I appreciate the interest and insights--


Anonymous said...

Hi Brett,
I ran the same 4 criteria on the S&P CNX Nifty (the National Stock Exchange of India Index) for the period 1996 to date (2732 trading days).

There are 16 instances of bearish engulfing patterns. Of the 16 data points, on the next day, the market closed up on 10 occasions, and down on 6. The next day returned an average of 0.36% (max. 2.79%, min. -1.97%).

Of the 2732 trading days, 1459 days the market closed up, and 1268 days closed down, for an average daily gain of 0.07%.

The same as your findings.


Brett Steenbarger, Ph.D. said...

Thanks for that interesting look, Brian. Many patterns that are commonly thought to be bearish end up not having a distinctive downside edge and vice versa--especially since the markets have shown more of a tendency toward mean reversion on short time frames. It would be interesting to test some of these patterns on commodities and currencies.