The key to drawing useful information out of noisy data is to rely on multiple “sensors.” Alone, each sensor may capture only a small portion of the true signal, and may not be greatly useful in and of itself. The power comes when the sensors are used together in order to distinguish the common signal of interest from the surrounding noise. For example, a good physician diagnoses a patient by collecting a history and a set of observations from various sensors (tests, measurements, symptoms, and so forth), looking for various “signatures” or combinations of evidence that are consistent with one specific condition or another. More generally, a "signature" is any distinctive combination of features that is associated with a specific identity, condition, or outcome.
The reason that doctors wince when people self-diagnose using the internet is that people tend to focus on individual symptoms rather than looking for complete, well-informed signatures. A single symptom, like headache, can be associated with dozens of conditions, so while it may be an informative piece of evidence, it’s not enough. A single symptom is usually a weak predictor in and of itself, and becomes useful only in the context of other evidence. Because patients focus on isolated symptoms rather than syndromes, doctors regularly receive frantic calls from patients that have self-diagnosed a brain tumor. Yes, a tiny fraction of those patients may actually have a brain tumor, but that diagnosis is confirmed only on the basis of a whole syndrome of other findings.
In the financial markets, useful sensors include valuations, price action, overbought/oversold measures, breadth, leadership, sentiment, credit market behavior, volatility, economic factors, and many others. Prior to the recent half-cycle advance since 2009, our own reliance on “signatures” drawn from multiple sensors allowed us to correctly anticipate the 2000-2002 and 2007-2009 collapses (with accurate loss estimates for both), and the evidence to shift to a constructive market outlook after every bear market decline over more than three decades.
As I've detailed regularly, the emergence of an extreme “overvalued, overbought, overbullish” syndrome of market conditions has historically been a powerful warning against further speculation. In prior cycles across history, the emergence of this syndrome was regularly accompanied or closely followed by deteriorating market internals (an indication of subtle underlying risk-aversion among investors), and a steep market loss would typically follow. In the half-cycle since 2009, our reliance on that aspect of history proved costly, because Fed-induced yield-seeking (“there is no alternative!”) encouraged investors to speculate long after severe “overvalued, overbought, overbullish” syndromes emerged. In mid-2014, we adapted by imposing an additional constraint on our methods: in the presence of zero-interest rates, one had to wait for market internals to deteriorate explicitly before adopting a hard-negative market outlook.
