How the DoubleTrends™ Regime Indicator is constructed

Most declines in the S&P 500 are routine: rebalancing, withdrawals and ordinary profit-taking that pass without much consequence. A few are not. They are acute, disorderly selloffs or long, macro-driven bear markets — the stretches where buying a decline has historically paid off, and where it is hardest to actually do.

The indicator tells the two apart. It first classifies the market’s state from price, volatility and macroeconomic inputs, and only then considers whether an oversold reversal is meaningful. The question it answers is narrow on purpose: is this a stressed market turning, or noise?

Two outputs are published each trading day: the regime (calm, correction, panic or bear) and the signal state. Signals are rare — 28 since Aug 1998, about 9.1 per decade — and some years have none.

The method uses five public series. There’s no proprietary or survey data involved.

3.1 Dual-horizon momentum

Williams %R — a 1970s momentum oscillator that locates the latest close within its recent high–low range — is computed over a short and a long lookback, and each is smoothed. A single horizon is either too noisy (short) or too slow (long); requiring both to agree weeds out most of the false readings. A market is oversold when both smoothed readings sit deep in the bottom of their ranges, and a reversal candidate occurs on the session in which that condition ends.

A momentum reading by itself isn’t treated as an edge. Our open-source AlphaVerify framework tests single market conditions against synthetic markets that contain no signal, and widely followed ones — the golden cross, price above its 200-day average — do not clear that test on the NASDAQ. Which is why a reversal candidate only gets considered inside a stress regime (sections 4 and 5).

3.2 The shape of a decline

Four measures describe what a decline is doing, not just how far it has gone: depth (distance below the trailing one-year closing high), duration (sessions since that high), sharpness (the worst ten-session move within the trailing month) and a reference high (the recent ceiling the rest is measured against). Together they distinguish a 3% drift over six weeks from a 15% fall in four.

3.3 Volatility and macro context

The VIX is used as a smoothed percentile of its own trailing year, so “high” is judged against what the VIX has done lately instead of a fixed level. The federal funds rate is read as its change over three and six months; unemployment as its six-month change and its position against its own trailing average; CPI as year-over-year inflation.

Each session receives exactly one regime. A base regime is decided from price and volatility alone, and a bear overlay can replace it.

4.1 Base regime

• Calm when implied volatility is in the low part of its trailing-year range.
• Panic when the index is below its recent high and either its recent ten-session decline is severe or implied volatility is near the top of its trailing-year range.
• Correction otherwise.

4.2 Bear overlay

The Bear regime marks a deep and persistent decline under macroeconomic pressure. It begins only when all of the following hold: the drawdown has lasted well beyond a typical correction; it is deep; the market has re-entered oversold territory repeatedly; and at least one source of macro pressure is present — rapid policy tightening, rising unemployment, or high inflation met by tightening. Once active it persists, overriding the base regime, until the drawdown has recovered materially.

Since December 1995 the overlay has been active three times: 2001–03, 2008–09 and 2022–23.

A signal fires on a session only when three conditions are met on that session.

5.1 Clusters

Reversals during one episode of stress are grouped into a cluster. Bear clusters stay open until the decline has recovered, because bear markets throw off plenty of short-lived rallies before they end. That’s why a bear-regime signal needs both depth of sequence and confirmation.

5.2 Signal window

A signal remains active for five sessions after it fires, and the signal date is recorded with it. Separately, subscribers receive an oversold alert when a panic or bear market is oversold but no reversal has yet occurred.

Readings and signals are back-tested unless they were published on the day: they were produced after the fact by applying the method to history, not published at the time. For prices and volatility the reconstruction is point-in-time — on each historical date the method sees only closes on or before that date.

Before any history goes out it is computed twice, separately: once over the full series and once by replaying each date separately with only the trailing window available on that date. The two must agree field by field on a random sample of dates and produce byte-identical output; otherwise nothing is published.

Macroeconomic timing. Monthly series are aligned to the month they describe, so in the back-tested history a month’s value is applied from the first day of that month — between one and six weeks before it was actually released. Live readings use only values already published. These series enter only the bear overlay (section 4.2), so the effect is limited to the timing of bear-regime episodes.

Data vintage. Macroeconomic series are revised after first release. The reconstruction uses the series as currently published, which can differ from the values available at the time. Price and volatility inputs are not revised.

• Signals can be early. In long declines the market can keep falling after a signal. The October 2000, September 2001 and January 2008 signals were each followed by lower prices a year later; the full record is on the historical data page.
• Rarity. Several years contain no signal. The indicator is not a source of regular trading activity.
• Scope. The method is calibrated to the S&P 500 only. Returns quoted on this site are index price returns, excluding dividends, costs and taxes; the index cannot be invested in directly.
• Not advice. The indicator describes market conditions. It does not account for any investor’s circumstances and is not a recommendation to buy or sell any security.