Source: Titman, S., Wei, K. C. J. & Xie, F. (2004). Journal of Financial and Quantitative Analysis 39(4), 677–700. (NBER WP 9951, 2003.) DOI: 10.1017/S0022109000003173
TL;DR
Firms that substantially increase capital investment subsequently earn negative benchmark-adjusted returns. The effect is strongest where managers have the most investment discretion — high cash flow and low debt — and is significant mainly in periods when hostile takeovers were less prevalent. Consistent with investors underreacting to the empire-building implications of high investment. The relation is independent of the long-term-reversal and equity-issuance anomalies.
What anomaly it documents
Predictor: abnormal capital investment (CI), a firm's current capex relative to its own recent trend.
Shape: monotone across CI-sorted portfolios; a low-minus-high (CI-spread) zero-cost portfolio earns positive returns.
A close cousin of the asset-growth (Cooper-Gulen-Schill) and accruals (Sloan) anomalies.
How to construct it
Signal: CI_{t-1} = CE_{t-1} / [(CE_{t-2} + CE_{t-3} + CE_{t-4})/3] − 1, where CE = capital expenditures (Compustat item 128) scaled by sales. CI = 0 means current investment equals the prior-3-year average; CI > 0 means above-trend ("high investor"). Sales is the deflator (benchmark assumed to grow with sales).
Form CI portfolios in formation year t and match returns from July of year t to June of year t+1 against CI_{t-1}.
Interact with managerial-discretion proxies: free cash flow (high CF) and leverage (low debt/assets).
Robustness measures include CE_{t-1} − (CE_{t-2}+CE_{t-3}+CE_{t-4})/3, CE_{t-1} alone, and a five-year benchmark.
Evidence and replication
Sample: Compustat financial data 1969–1995; return test period July 1973–June 1996.
The low-minus-high CI-spread mean excess return is 0.168%/month (t = 2.91), ≈ 2.02%/year; positive in 15 of 17 years (test statistic 4.75, rejects the null).
The effect concentrates among high-cash-flow firms (CI-spread 0.227%/mo vs 0.078% for low CF) and low-debt firms (0.225%/mo, significant) — those most able to over-invest without discipline.
It is significant only in periods with fewer hostile takeovers (0.312%/month, t = 4.42 in non-hostile-takeover years vs 0.192% all years), supporting the agency interpretation.
Why it might work
Agency / over-investment (empire building): managers expand beyond value-maximizing levels (Jensen 1986) and oversell prospects when raising capital; the market underreacts and later corrects.
q-theory / risk-based: firms invest more when discount rates (expected returns) are low, mechanically lowering future returns.
Limitations and risks
Overlaps with asset growth and accruals; isolating the pure capex channel is difficult.
Accounting-based and annual (June rebalance); transaction costs, data timing, and small spreads (~0.2%/month) matter for implementation.
The takeover-period dependence suggests the premium can vary with the corporate-governance regime.
Key references
Titman, S., Wei, K. C. & Xie, F. (2004) — Capital Investments and Stock Returns — JFQA
Cooper, M., Gulen, H. & Schill, M. (2008) — Asset Growth and the Cross-Section of Stock Returns — Journal of Finance
Sloan, R. (1996) — Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? — The Accounting Review
Provenance: verified/generated from the paper's full text.
