Paper Trail #31: Capital Investments and Stock Returns (Titman, Wei & Xie, 2004) — the direct primary source of the CMA investment factor in FF 2015 (PT #22), on a 58,880 firm-year sample the paper documents a CI-spread of 0.168% per month that is positive in 15 of 17 non-hostile-takeover years and NEGATIVE in every one of the 1984-1989 hostile-takeover years

Titman, Wei & Xie (2004) — Capital Investments and Stock Returns, Journal of Financial and Quantitative Analysis 39(4) pp. 677-700. Firms that substantially increase capital investments subsequently earn negative benchmark-adjusted returns. 58,880 industrial firm-years, Compustat + CRSP, Jul 1973–Jun 1996 (23-year sample). Long lowest-two-CI quintiles / short highest-two = 0.168% per month excess return, positive in 15 of 17 non-hostile-takeover years and NEGATIVE in every one of the 1984-1989 hostile-takeover years.

Direct primary source of the CMA (Conservative-Minus-Aggressive) investment factor in Fama-French 2015 (PT #22) and of the asset-growth anomaly that Cooper-Gulen-Schill 2008 (PT #26) extends to the full firm universe. Full primary source verified via NBER working paper 9951.

The CI measure

Section III.B of the paper defines the capital-investment measure:

CI_{t-1} = CE_{t-1} / [(CE_{t-2} + CE_{t-3} + CE_{t-4}) / 3] − 1

Where CE_{t-1}= firm’s capital expenditures (Compustat data item 128) in fiscal year t-1 scaled by its sales in year t-1. Interpretation: CI = 0 means the firm invested at its own trailing 3-year capex-to-sales average; CI > 0 means it invested more than trend; CI < 0 means less than trend. Assumes benchmark capital expenditures grow proportionally with sales.

Formation: annually at end of June, all industrial firms (utilities, REITs and financials excluded) with ≥ 2 years on Compustat get sorted into 5 CI quintiles. Value-weighted excess returns are computed from July of year t to June of year t+1, rebalanced each year. Sample: 58,880 firm-years over 1969-1995 financial data / Jul 1973 – Jun 1996 return period, averaging 2,560 firms per year (reduced to 1,725 after the 5-year CAPX-history filter).

The 1984-1989 hostile-takeover natural experiment

Panel B of Table 1 reports the year-by-year CI-spread performance Jul 1973 – Jun 1996. In non-hostile-takeover years (17 total across 1973-1983 and 1990-1996), the CI-spread is POSITIVE in 15 of 17 years. In hostile-takeover years (1984-1989, 6 years), the CI-spread is NEGATIVE in EVERY year.

Mechanism: during the 1980s hostile-takeover era, managers who over-invested faced credible takeover risk that disciplined the empire-building tendency. The negative CI-return relation only emerges when internal discipline (leverage, cash-flow scrutiny) is weak AND external (takeover) discipline is absent. This is the paper’s cleanest identification argument — the anomaly is causally about empire building rather than a data-mining coincidence, because the mechanism’s off-switch (hostile takeover pressure) turns the anomaly off in exactly the years the mechanism predicts it should.

The Jensen (1986) free-cash-flow interaction

Jensen (1986) "free cash flow" theory predicts empire-building is worst when internal cash is high (no external capital constraint) AND leverage is low (no debt-service discipline against pet-project investment). TWX 2004 tests this directly by double-sorting on cash-flow tercile × CI quintile:

Combined high-cash-flow / low-debt subsample: CI-spread = 0.461% per month — 2.75x the whole-sample 0.168%. Combined low-cash-flow / high-debt subsample: CI-spread near zero, statistically insignificant. The anomaly concentrates in exactly the empire-building-permissive subsample the theory predicts, and disappears in the empire- building-constrained subsample.

What survives risk-factor adjustment

TWX 2004 runs the 5 CI-portfolio returns through the Carhart four-factor model (Fama-French 3 factors + momentum). The intercepts (four-factor alphas) are significantly different across the 5 portfolios at Wilks’ Lambda p < 0.01. The CI-spread alpha remains significant AFTER controlling for market, size, book-to-market, and momentum.

The abstract explicitly closes two other candidate explanations: the CI-return relation is independent of long-term return reversal (DeBondt-Thaler 1985) and independent of secondary equity issue anomalies. The CI signal is a genuinely distinct anomaly, not a rediscovery of one of the previously documented anomalies TWX 2004 explicitly controls for.

Earnings-announcement concentration

Panel A of Table 3 examines the timing of the abnormal return within the first post-formation year. A disproportionate share of the annual abnormal return concentrates in the ±3 trading days around each of the 4 quarterly earnings announcements — the ratio of announcement-window-return to full-year-return is roughly 30%, despite the announcement windows being only 12 of 252 trading days per year (4.8%).

Interpretation: investors gradually learn that a high-CI firm is over-investing, and the correction happens preferentially when new earnings information forces the update. This connects TWX 2004 to the post-earnings-announcement-drift literature covered in Bernard-Thomas 1989 (PT #6) and Ball-Brown 1968 (PT #15): investors respond to information gradually and asymmetrically around scheduled information events, not smoothly.

Downstream lineage

Cooper, Gulen & Schill 2008 (PT #26). Extends TWX 2004’s CI signal from capex-based measure to asset-growth-based measure (Δtotal assets / total assets). CGS catches investment discretion beyond just capital expenditures (M&A, working capital, R&D). CGS §I explicitly frames itself as "strengthening and generalising" TWX 2004.

Fama-French 2015 5-factor (PT #22).Adds a CMA (Conservative-Minus-Aggressive) investment factor to the 3-factor model. §II.C acknowledges TWX 2004 as one of two direct primary sources for the investment signal (Anderson & Garcia-Feijóo 2006 the other). FF 2015 uses CGS’s asset-growth measure over TWX’s CI measure because CGS covers a broader firm universe; but the empire-building intuition traces back to TWX 2004.

Novy-Marx 2013 (PT #24).Complementary to TWX 2004: Novy-Marx documents the gross profitability signal that FF 2015 RMW factor is built on; TWX 2004 documents the CI signal that CMA is built on. RMW + CMA together are the two "quality" additions to the FF 3-factor model.

Piotroski 2000 (PT #28).F-score contains an F_ΔTURN signal (asset turnover change YoY) that is inversely related to TWX 2004’s CI measure — firms increasing capex without corresponding sales growth exhibit falling asset turnover.

Verification note

Full primary source verified via WebFetch + pymupdf on 2026-08-31 from NBER Working Paper 9951 (September 2003), the pre-publication draft matching the JFQA 2004 published version. 40 pages, 279.9 KB, 92,071 characters text-native — no OCR required. 23rd Paper Trail post out of 31 with full primary-source access. Every abstract line, sample statistic, headline return figure, formula, and "15 of 17 years" claim traces to a specific verbatim passage in the PDF. The published JFQA version is behind Cambridge Core’s paywall; the NBER draft (nber.org/papers/w9951) is the accessible primary source.

1-day spacing from PT #30 (Cochrane 2011) justified by different lineage — Cochrane was asset-pricing meta-framework; TWX 2004 is a specific-anomaly primary source that closes the ledger’s 5-day-held queue item flagging "the investment-anomaly primary source that PT #26 (CGS 2008) builds on."