Tuesday, March 11, 2008

"Positive" Divergence of New Lows?

One statistical divergence I’ve seen some discussion of lately is the smaller number of new lows compared to the January bottom. Theory says that this is a positive breadth indication. Since less stocks are posting new lows, less stocks are in poor technical shape. Hence, although the price level of the observed index is near or below the previous swing low, the makeup of the market is improved. Supposedly this has bullish connotations looking forward.

Proponents of this kind of analysis can easily point to some instances where the divergence seemed to work beautifully. One nice looking example would be August 2004 bottom. The S&P 500 poked beneath the May lows but NYSE New Lows contracted. The market put in a nice rally after that.

I ran a test to see if a contraction of new lows on a swing lower for the S&P 500 was predictive of a rally. Basically I looked for the SPX to make a 100 day low while the highest number of new lows in the last 100 days was greater than the highest number of new lows in the last 10 days. The trade entry point for the study was above the prior days high and the exit was 20 days later. Going back to 1992 I found 10 instances. I’ve listed them below.






It appears to me this divergence worked well during bull markets (98, 99, 04, 06) and not well during the bear market of 2000 – 2002. Success would therefore seem to be attributable to factors other than the divergence.

October ’98 and October ’02 launched some very strong 1-month moves and it’s interesting that the divergence was in place at those times. Based on the magnitude of success of some of these rallies the divergence may therefore be notable. As a stand alone indicator I was unable to find predictive value using my fairly simple test.

1 comment:

Tim said...

Can you test something along this line:

Ratio of stocks up 50%+ in 1 month vs stocks down 50% or more in 1 month. Seen on another site that when ratio is 2 to 1 stocks up then potentially overbought, reverse is true.

Thanks.