The smooth fit says maybe not.
The split in half says yes, modestly.

Backlog #20: do buybacks boost returns? Share buybacks get pitched by executives as capital discipline — returning excess cash to shareholders at a price management believes undervalues the stock — and attacked by critics as financial engineering that inflates per-share metrics without creating anything. Both stories make the same testable prediction: a company spending more of its own market cap buying back stock should, on average, go on to do better. This run pulls the SEC's own structured disclosure of cash actually spent on buybacks (us-gaap:PaymentsForRepurchaseOfCommonStock) for every company that reported a clean full-calendar-year 2022 figure, divides by each company's market cap at the end of 2022 (SEC-reported shares outstanding × year-end price), and regresses that buyback yield against the stock's CY2023 forward total return — 405 real companies, common stock only, funds and trusts excluded.

The straightforward regression almost clears this desk's bar, and then doesn't once you look at what's driving it. Naive OLS: slope +0.57, 95% CI [-0.02, +1.17], p=0.059 — technically still contains zero, and barely. Drop just the single most extreme forward return in the sample — CASI Pharmaceuticals, +300%, a penny stock — and the fit falls apart: slope +0.41, CI [-0.15, +0.96], p=0.15. An HC3-robust standard error, which doesn't pretend the variance is uniform across the sample, agrees from the start: CI [-0.26, +1.40], p=0.18. The naive near-miss is one stock's arithmetic, not a relationship.

A measure that doesn't care how extreme the biggest winners are tells a different, smaller story. Spearman's rank correlation — does higher buyback yield simply track with a higher-ranked forward return, regardless of magnitude — comes back ρ=0.153, p=0.0020, and barely moves after dropping the same three most extreme returns one at a time (ρ stays between 0.142 and 0.153, p never rises above 0.0044). There is a real, if weak, monotonic ordering here that a straight line fitted to a right-skewed, outlier-prone return distribution is the wrong tool to see cleanly.

Splitting the same 405 companies at the median buyback yield into heavy vs. light buybackers makes the rank-based signal concrete. The 203 heaviest buybackers averaged +21.4% the following year; the 202 lightest averaged +13.4% — a gap of +8.0 points, Welch's p=0.030, and a 4,000-draw bootstrap on the gap returns 95% CI [+0.8, +14.9] — excludes zero. That gap is noisier than the bootstrap CI alone suggests: dropping the same one, two, or three most extreme returns moves the gap between +5.2 and +8.0 points and its Welch p-value between 0.030 and 0.111 — crossing back above 0.05 when the second-most-extreme point is removed. Reported as a real but fragile split, not a precision estimate.

The one check built specifically to catch a confound says the relationship, weak as it already was, does not survive it. Smaller, cheaper-looking companies both buy back more of their own stock (as a share of market cap) and were exactly the part of the market that led the 2023 rebound. Controlling for company size (log market cap) alongside buyback yield: the buyback-yield slope drops to +0.51, CI [-0.29, +1.31], p=0.21 — contains zero. A log(1+return) transform, which shrinks exactly the right tail driving the naive near-miss, agrees: CI [-0.22, +1.23].