One of my works in progress is tracking the number of NYSE stocks each day that give buy signals and sell signals across different technical indicators. Above are charts for NYSE stocks giving buy vs. sell signals for Bollinger Bands (top chart) and the Commodity Channel Index (bottom chart). (Raw data from the Stock Charts site). As a rule, we see the number of buy signals peak ahead of price during a market cycle and the number of sell signals anticipate a cycle price low. Sell signals for the current cycle peaked in mid-January, with fewer sell signals posted at the late January/early February lows. Buy signals for the current cycle peaked early in February.
One interesting facet of the Bollinger measure is that it is the absence of weakness--and not just the presence of strength--that alerts us to a strong stock market. When the market is ready to turn over, there are typically weak shares and sectors leading the way and that shows up as a relatively elevated number of sell signals for the Bollinger Band measure, even as the index price has been near highs. On the other hand, when very few stocks and sectors are weak, the market often drifts higher, as selling pressure is minimal.
For example, I looked at the period from early May, 2014 (when I first began assembling these data) to the present and broke down the number of sell signals for the Bollinger Band measure in a simple median split. When we had few sell signals, the next five days in SPY rose by an average of +.37%. When we had more sell signals, the next five days in SPY rose by an average of only +.06%.
Interestingly, daily sell signals for the Bollinger Band measure correlate only +.07 with the RSI measure and only +.21 with the MACD measure. The absence of sell signals for those latter two indicators has not led to superior returns going forward. Indeed, when we've had few MACD sell signals, the next five days in SPY have averaged a gain of only +.09% vs. +.34% when we've had many sell signals.
It appears that separating the number of buy and sell signals and looking within each indicator captures different time frames and different patterns of momentum and reversal. This strikes me as a most promising area of research, particularly when we focus on technical indicators that are not highly correlated.
Further Reading: Technical Indicators From the Bottom Up