Macro Inflection Points

What defines an inflection point?

It's the output of an algorithm called PELT (Pruned Exact Linear Time), running a statistical model called RBF (radial basis function cost) on each series. In plain terms: it scans the whole time series and finds the dates where the statistical character of the data changed significantly — a shift in level, variance, or trend that's large enough to not be just noise. It's not "the line moved a lot" — it's closer to "the data before this point and the data after this point look like they came from different regimes."

A tunable penalty parameter controls sensitivity (lower penalty = more, smaller inflection points get flagged; higher = only major regime shifts). Each series in this app has its own tuned penalty, since a Treasury yield and a jobs count behave very differently.

Why do some rows show no matched event?

This is a separate step from detection. After PELT finds an inflection point, a second step checks: is there a historical event in the curated list of 53 events within 45 days of this date? If yes, it fills in the event. If no, it leaves it blank.

So a blank row means: the algorithm found a real, statistically significant shift in the data — but nothing in the hand-picked event list happened near that date. That's expected and honest, not a bug — three real reasons this happens:

1. The cause is real but isn't one of the 53 curated events (the list is deliberately a "greatest hits" of wars, elections, and crises, not exhaustive — e.g. a Fed rate decision that isn't in the list, a data-release quirk, a smaller regional event)
2. It's a slower-building trend shift with no single triggering headline
3. It could be a false positive from the algorithm — PELT is tuned per-series but isn't perfect