How late in the rally are we?
The regime classifier answers a fast question: is risk appetite on right now? These four gauges answer a slow one: how mature is the cycle that's producing that risk appetite? They move on monthly and quarterly data, so they don't belong in a daily composite directly — but a loud risk-on composite combined with saturated late-cycle gauges is historically the configuration where trimming into strength beats chasing.
The selection is inspired by BofA's “signposts of a market peak” checklist (10 signals, ~70% triggered as of May 2026, vs 80–90% at the 1990/2000/2007 peaks) — rebuilt here from free sources only. The honest caveat up front: peak checklists are assembled retrospectively, every signal is in-sample by construction, and the 2018/2020 rows of BofA's own table show the false-positive cost. Read these as gradients, not triggers.
1. Index concentration — the top of the index vs the index
Cap-weighted S&P 500 (SPY) divided by equal-weighted (RSP). When a handful of megacaps carry the index, cap-weight outruns equal-weight and the ratio climbs. This is the generic version of “top 7 vs the index” — it doesn't hard-code today's leaders, so it keeps working when leadership rotates. We chart the top-7-today basket as a secondary line; its past values carry a survivorship caveat (we picked today's winners), which is exactly why the generic ratio is the primary signal. Extreme concentration is not a sell signal by itself — it can persist for years — but every major unwind (2000, 2022) started from a concentration extreme, because an index priced off 7 stories has fewer independent legs to stand on.
2. M&A activity — S-4 filings per month
Monthly count of S-4 registration statements filed with the SEC (the form companies file for stock-financed mergers and exchange offers), via EDGAR full-text search. Deal-making is a confidence trade: boards commit capital at scale when financing is cheap and projections extrapolate well — which is why deal counts historically crest near cycle tops (BofA's version of this trigger: 5y deal-count Z-score > 1). Mechanical, free, and harder to game than survey data.
3. Margin debt — speculative leverage
Broker-dealer margin loans from the Fed's Z.1 flow-of-funds accounts (quarterly), read as year-over-year growth. This is the liquidity-and-leverage proxy: when investors borrow aggressively against portfolios, marginal buying power is being manufactured rather than earned. YoY growth peaks led the 2000, 2007 and 2021 equity peaks; the deepest contractions marked durable bottoms. Quarterly data means it's slow — treat it as cycle positioning, not timing.
4. Consumer confidence — the contrarian survey
University of Michigan consumer sentiment (the free cousin of the Conference Board series BofA uses with its >110 trigger; scales differ, shape is similar). Confidence is a coincident-to-contrarian signal: very high readings mean the good news is fully absorbed — households are all-in — which is late-cycle. The classic pre-recession pattern is elevated-then-rolling-over. Very low readings are historically excellent forward-return setups (2009, 2022).
How these would feed the regime composite
Proposed panel assignments if promoted: concentration, M&A activity and margin debt join the equity factor panel; consumer confidence joins the macro panel. All four would be sign-aligned the same way as existing indicators (percentile-ranked, +1 = risk-on), with concentration and M&A carrying a negative sign at the extremes — the subtlety being that these are hump-shaped signals: rising from low levels is healthy risk-on, saturated readings are late-cycle. That nonlinearity is why they start here as a research preview rather than going straight into a linear panel, and why any promotion would be a version bump with a documented relock, per data/regime/regime-versions.json.
Read together with the channel-divergence preview: if late-cycle gauges saturate in equities while crypto isn't absorbing risk-appetite flow, that's the “AI vs crypto” configuration — equity euphoria without crypto beta — and the composite should read it as something other than uniform risk-on.
Back to /regime.