"This market is just like 1994." Or 1999. Or 1929, or ....
These metaphors and comparisons are commonplace, but dangerous. Subjective similarities are often framed in a single dimension, or (worse) are used to infer expected outcomes from a single prior case.
DejaVu is different. Similarities are explicit and quantified, and historical parallels are ruthlessly objective. Prior occasions are identified by specific date, and their outcomes are compiled over explicit time horizons. For acceptable inference, precedents must be 1) sizeable, 2) numerous, and 3) consistent.
Those three elements lie at the heart of statistical significance. When all three are (jointly) sufficient, we have confidence the outcomes are more than coincidence. And we can hold that confidence to any threshold of certainty we demand.
Some novel metrics have emerged from DejaVu. But the model is also useful exploring market data that we think we already know pretty well. Here are two links to DejaVu analysis applied to well-known market indicators....
DejaVu is useful both for testing specific constructs (like the Fed and VIX analyses above), and also for continuous inference from daily market data (like our Daily States tabulations).
DejaVu makes historical analysis efficient and productive. The model provides an effective way to test and quantify actual market effects. Historic relationships are validated and up-dated, often revealing new and emerging (or decaying) relationships. Time horizons and response cycles are assessed with rigor, including expected values and probabilities. In effect, DejaVu offers statistically rigorous "studies," on a ready on-call basis.
Tactical performance indicators grow out of DejaVu studies, sometimes discovering latent relationships not easily discerned by eye. The PDQ filter, for instance, is a process with productive capacity more than 2x S&P, with only 60% of the volatility. But the key PDQ input ("NYSE Fluidity") is not a successful directional metric overall; it is selective and occasional. Overall correlations and tabulations would miss the point entirely.
In continuous screening applications, current market metrics are screened back through history for prior occasions "like" today. Often the matches are too sparse or too diverse for inference. (Much of the time, after all, market conditions are unremarkable.) But sometimes--when prior occasions are numerous, and outcomes are sizeable and consistent--the odds of similar outcome climb dramatically. Just how numerous, how large, and how consistent is resolved jointly in significance testing.
For examples of value-added market indicators that have emerged from DejaVu analysis, check our pages on Performance. For a summary of recent DejaVu market screens, go to Market States.
To enquire about DejaVu applications with your own data and indicators, you can leave a message back at the Models page, or use the master checklist of Analytix Services. Or call directly anytime, at (603) 643-6430.