The Sum of All FEARS: Investor Sentiment and Asset Prices
Source: Da, Z., Engelberg, J. & Gao, P. (2015) · Review of Financial Studies 28(1), 1–32 · DOI: 10.1093/rfs/hhu072
TL;DR
A method for measuring market-level investor sentiment directly from household Internet search behavior. Aggregating daily Google search volume for anxiety-related economic terms (e.g., "recession," "unemployment," "bankruptcy") yields the FEARS index (Financial and Economic Attitudes Revealed by Search). Over January 2004–December 2011, FEARS predicts short-term return reversals, temporary increases in volatility, and mutual-fund flows out of equity funds into bond funds — broadly consistent with noise-trader (DSSW) sentiment theory.
Problem it solves
Sentiment is central to behavioral asset pricing but hard to measure. Market-based proxies (closed-end fund discount, IPO volume, VIX, fund flows) are equilibrium outcomes contaminated by other forces ("how do you test inputs→outputs with an output measure?"), and survey indices are low-frequency. FEARS offers a high-frequency, behavior-revealed, input-side sentiment measure.
The method
Primitive word list. Start from the Harvard IV-4 and Lasswell Value dictionaries to assemble a "primitive" list (149 primitive words).
Search-term expansion. Enter each primitive into Google Trends, which returns ten "top searches" related to it; filter out irrelevant terms and those with too few valid Search Volume Index (SVI) values. This yields 118 economically meaningful search terms.
Abnormal search volume. Download daily SVI for each of the 118 terms (restricted to U.S.), take daily log differences, winsorize, and remove intra-week/intra-year seasonality to get abnormal search volume (ASVI) for each term.
Dynamic aggregation. Rather than fix weights, run a rolling regression of market returns on each term's ASVI over the prior history and keep only terms historically related to returns. This produces a dynamic list of 30 search terms whose ASVI are aggregated into the daily FEARS index. (One s.d. of FEARS ≈ 0.3549.)
Assumptions & inputs
Inputs: Google Trends SVI for the 118 terms, the dictionary-seeded term list, daily market data. Output: a single daily FEARS time series.
Assumes search queries reveal genuine household economic anxiety (sentiment), and that aggregating anxiety searches captures a sentiment shock distinct from fundamentals; controls include VIX, EPU, and the ADS business-conditions index.
How to use it
Use FEARS as a daily sentiment regressor. Headline results (Table 2, S&P 500, controlling for lagged returns, VIX, EPU, ADS): a one-s.d. rise in FEARS coincides with a −19 bps same-day (day 0) return, then +7.1 bps at k=1 (5%) and +7.3 bps at k=2 (10%), a cumulative +14.4 bps over days 1–2 (1%) — i.e., the day-0 move is almost fully reversed within two days, with no significant effect at k=3–5.
The reversal is strongest among high-beta, high-volatility, harder-to-arbitrage stocks; similar spike-reversal patterns appear in other asset classes and futures.
Limitations & pitfalls
Search-term selection and signing involve judgment; Google Trends data are sampled/revised and U.S.-geographically restricted here.
Effects are short-horizon and reverse, and overlap with liquidity-shock (Campbell–Grossman–Wang) interpretations; transaction costs limit exploitation.
Terms are chosen partly on historical market correlation, so out-of-sample predictive power should be interpreted with that selection in mind.
Key references
Da, Z., Engelberg, J. & Gao, P. (2015) — The Sum of All FEARS — Review of Financial Studies
Da, Z., Engelberg, J. & Gao, P. (2011) — In Search of Attention — Journal of Finance
De Long, Shleifer, Summers & Waldmann (1990) — Noise Trader Risk in Financial Markets — Journal of Political Economy
Baker, M. & Wurgler, J. (2006) — Investor Sentiment and the Cross-Section of Stock Returns — Journal of Finance
Provenance: verified/generated from the paper's full text.
