Saturday, September 01, 2007

Rises and Declines on High and Low Volume: Test Before You Invest

I've been hearing a number of commentators speculate on the validity of market rises and declines based upon the volume of shares being traded that day. So, for example, a rise on low volume is discounted and even looked upon bearishly, presumably because the higher prices are not attracting greater participation.

What gives me pause is that I never see any efforts to quantify such commonly-held wisdom. It's part of the technical analysis lore, but is it valid? Does volume on a rise or decline affect the market's subsequent behavior?

I went back to the start of 1999 (N = 2178 trading days) in the S&P 500 Index (SPY) and identified all instances in which the market either rose more than 1% in a day (N = 320) or declined more than 1% in a day (N = 340). I then calculated each day's trading volume as a proportion of the prior 200 days' volume and conducted a median split of the data.

That means that we're looking at strong and weak market days on high or low relative volume.

When SPY has been up more than 1% in a day (N = 320), the next day averages a rise of .02% (164 up, 156 down). When the rise is on relatively high volume (N = 160), the next day averages a flat performance (80 up, 80 down). When the rise is on relatively low volume (N = 160), the next day averages a gain of .04% (84 up, 76 down). Clearly, there's no general indication that a rise on high volume is any more bullish than one on low volume.

If we just look at those occasions in which SPY has been up more than 1% and volume has been twice (or more) the 200-day average (N = 37), the next day in SPY averages a gain of .06% (23 up, 14 down). This is a slight bullish edge in a limited set of circumstances.

If we limit our look to those occasions in which SPY has been up more than 1% and volume has been less than 80% of the 200-day average (N = 53), the next day in SPY has averaged a gain of .08% (29 up, 24 down). Low volume has not led to inferior next day returns for rising days.

Conversely, when SPY has been down more than 1% in a day (N = 340), the next day averages a rise of .15% (191 up, 149 down). When the drop in SPY is on relatively high volume (N = 170), the next day averages a gain of .16% (101 up, 69 down). When the drop in SPY is on relatively low volume (N = 170), the next day averages a gain of .14% (90 up, 80 down). Again, volume plays a very minor role in determining next day outcomes.

When we limit our look to those occasions in which SPY has been down more than 1% and volume has been twice (or more) the 200-day average (N = 57), the next day in SPY averages a gain of .37% (35 up, 22 down). That is a nice bullish edge, again in a limited set of circumstances.

Finally, if we examine occasions in which SPY has been down more than 1% and volume has been less than 80% of the 200-day average (N = 33), the next day in SPY has averaged a gain of .46% (19 up, 14 down). Interestingly, very low volume has been as bullish for declining days as very high volume.

The bottom line, it seems to me, is that greatly expanded volume on a rise or decline may be associated with better returns, but there is no evidence that low volume is followed by weaker daily returns. Indeed, when volume has been very low on rising and falling days, returns have tended, if anything, to be superior. Moderately elevated volume appears to have no significant impact on returns whatsoever.

Moreover, volume as a whole seems to play less of a role in next day returns than the simple fact of whether the prior day rose or fell by 1%. Returns were better following 1% declining days than 1% rising days regardless of market volume.

One has to wonder why traders would follow untested assumptions in putting their capital at risk. We're not talking rocket science here: my analysis required no programming and was conducted entirely in Excel in well under an hour. The moral of the story is to not accept market truisms on face value: Test before you invest.

RELEVANT POST:

Historical Patterns as a Heads Up in Trading
.

7 comments:

F-Trader said...

I think traders confuse the market propensity to reverse low volume range breakouts for a predictive indicator of trend. In my opinion, breadth and momentum are more predictive of short-term trend than relative volume figures.

I'm a big fan Dr. Steenbarger. Please stop by my blog ftrading.blogspot.com and say hi!

Aneesh said...

interesting. the "volume confirms strength of trend" heuristic is indeed one i've heard repeatedly. so i like the fact that brett is challening convention. but to make the study more complete, we would need to see the 1-week, 2-week, 10-week, etc. average changes in SPY after high- and low-volume moves. i don't know how much you can infer about the broader trend from just looking at the day after the move.

perhaps a more revealing question for the community to explore is the strength of the OBV (On Balance Volume) indicator, since this can quantify the relationship between volume and price over a longer period of time.

profste said...

i completely agree with aneesh, while reading your interesting post i hought immediately about the limitation of your test in being limited to the following day.

if we just think about the altenation between trend days and congestion days (as in taylor trading technique teached by lbr) after an up or down day we should expect a Z day (limited range).

i believe that an up day on low volume could show its bearish implications in the folowing days and the same consequence can't be found in down days as the market "could fall on its own weight" and doesn't necessarily need an increase in volume to confirm down days.

anyway this is just my empiric view, based on actual market observation. about the backtesting i completely agree with the latest james dalton which told that the only thing he would change in his first book would be to reduce the importance he gave to backtesting, as shown during this very last month of august, where the exceptions could lead to exceptional gains/losses.

regards, stefano

p.s. if you could consider modifying your Rss feed to full text my mobile connection monthly bill would be grateful :)
anyway thanks again for everything you do: it's a precious contribution to my growth

Ziad said...

Very interesting article. I personally have struggled with this concept of low vs. high volume moves, but I have looked at it more for intraday expectations rather than next day. For instance, if a morning decline is relatively sharp and coming on low volume, I have been apt to look for a reversal. However, several times I have gotten killed as the market just keeps following through even with the lack of volume. That would seem to support this study in principle, that volume is a poor predictor of subsequent market action on a short-term basis.

But so I have to ask, why have I held this notion to be true? Well, I think there are three reasons. One is that we usually associate low volume retracements as healthy and apt to reverse and extend the larger trend, so we extrapolate that principle and think that any low volume move is more apt to reverse than continue. Second, is that there is statistical evidence that low volume is associated with more range bound trade and less trending action, and so we start expecting a reversal when we see low volume moves. The third reason, as mentioned in the first comment above, is that we see that low volume breakouts usually reverse, and so again we generalize that principle to trend.

These all seem like logical deductions to make, but from limited experience I have noticed that they can get you in to trouble; many times the most persistent trends are low volume ones that keep on going with hardly any retracement. This then presents a certain dilemma: How can I associate low volume with range bound markets, and yet not expect low volume moves to be more likely to reverse than trend? Trying to answer this question could lead to a greater understanding of this tricky subject.

Brett Steenbarger, Ph.D. said...

Hi F-Trader,

Thanks for the comment, and thanks for the link to your blog!

Brett

Brett Steenbarger, Ph.D. said...

Hi Aneesh and Stefano,

Thanks for your comments; I will be looking at other time horizons as well!

Stefano, I've enabled full text for my blog; I'll keep it that way as long as it's not used to pirate my content and publish without my authorization. Huge problem. Thanks for the interest--

Brett

Brett Steenbarger, Ph.D. said...

Hi Ziad,

Yes, I do think there's worthwhile information in volume, just not statistical main effects. Volume does inform us about volatility, and it does provide useful information at range extremes to help us handicap the odds of breakout vs. mean reversion.

Brett