Measured unconditionally, an individual macro indicator's level carries almost no forward equity-return information: across 2000–2026 the largest absolute correlation with the forward 3-month S&P return is 0.20 (initial claims), and on the overlap-corrected effective sample it is barely distinguishable from zero — with the 'wrong', reaction-function sign. What sorts returns is the growth×inflation regime. Built on real-time (expanding-window) standardization to defeat look-ahead, the only robust claim is binary: stagflation is the one quadrant that doesn't pay. Non-stagflation regimes beat it by +3.4pp per quarter (Newey-West t = 2.5), stable across all thirteen robustness variants — while the four-way ranking is within noise (the bootstrap intervals overlap). A regime-driven allocation lifts the Sharpe from 0.54 to 0.68 and halves the drawdown (−32% vs −52%), almost entirely by de-risking the stagflation corner. The model reads reflation today, but at low confidence — growth is only +0.1σ into acceleration.

Return on equity is three different engines — margin, turnover, leverage — and the source matters more than the level. Decomposing 59 large-cap non-financials from their 10-K filings on average balances (2010–2025): the highest headline ROEs are largely a denominator effect, not financial leverage and not margin. For 11 names buybacks have shrunk book equity so far that ROE is mechanically meaningless — Home Depot's apparent '22× equity multiplier' is just 1.95× net-debt/EBITDA. The equity multiplier correlates only 0.24 with true net leverage, and a high-multiplier long-short portfolio carries no significant rate beta (t = 1.5) — refuting the claim that buyback-built ROE is rate-sensitive. A reproducible log-decomposition classifier and a formal quality score (NVIDIA tops it; net cash, 94% ROIC) replace hand-labels. The forward tests: turnover-sourced ROE has the most stable margins and the best subsequent returns, while — counterintuitively — leverage-sourced ROE is the most persistent, because a buyback policy is stickier than a margin competition erodes. And within this survivor basket, the decade's stock gains came more from revenue growth and multiple re-rating than from margins.

Decomposing daily nominal-yield changes into real and breakeven-inflation components, controlling for equity beta, and using Newey-West (HAC) standard errors overturns the 'Technology is long-duration' consensus. On real yields the rate-beta ladder is uniform and significant — every sector loads negatively — but the magnitude concentrates in bond proxies: Real Estate (−5.1% per +100bp, t=−8.1 after market control) and Utilities (−4.2%, t=−7.6). Technology's apparent rate sensitivity is almost entirely equity beta: with the market factor included its real-rate beta is +0.7% (t=1.8), and a five-factor model attributes 86% of its variance to market beta. The positive nominal betas of Energy (+3.6%) and Financials (+1.1%) are a breakeven channel (Energy breakeven β +18.0%, t=10.1), not a real-rate benefit. Growth-minus-value shows no robust real-rate duration at daily frequency across three definitions. Bond-proxy duration is structural across 2010–2026; Technology's is not.