Research
The Narrative Factor: A Systematic Approach to Capturing Narrative Alpha from Public Discourse
Date Written: April 30, 2026 | Revised: May 20, 2026
Abstract: Market-moving narratives drive equity returns yet remain largely absent from systematic factor models. This paper introduces a narrative factor built from ForecastOS Hivemind, which quantifies company-level exposures to 14 macro narratives spanning geopolitics, technology, and monetary policy. The composite factor dynamically weights active narratives -- those whose rolling 90-day relative discussion level is within 80% of its trailing six-month high *and* whose rolling 30-day relative discussion level is within 50% of its trailing six-month high -- by their share of total discussion volume, then standardizes cross-sectionally within industry (and subindustry) groups.
Tested in a long-short portfolio of the top 500 U.S. equities from January 2020 through April 2026, the factor-neutralized narrative portfolio -- constructed by pre-orthogonalizing the narrative signal against momentum, value, quality, leverage, and beta (cross-sectional residualization, then top/bottom decile construction) -- achieves an information ratio of **1.49x (industry-neutral)** and **0.88x (sub-industry-neutral)**, with maximum drawdown of 4.26% and 4.13% respectively. Under industry neutralization the portfolio achieves a positive IR in every calendar year of the sample. As a benchmark against which to read the narrative numbers, a 12-minus-1-month momentum factor constructed under the same factor-neutralization (residualizing momentum on value, quality, leverage, and beta) -- implemented with a monthly (21 trading day) full-reset rebalance, reflecting how momentum strategies are typically run in practice -- delivers an information ratio of 0.58x in both specifications with maximum drawdown of 22.73% (industry) and 25.03% (sub-industry). Cross-sectional correlation between the factor-neutralized narrative and momentum signals is zero by construction.
A Fama-MacBeth cross-sectional regression reinforces this at the signal level: the orthogonalized narrative characteristic produces t-statistics of **2.56 (industry-neutral) and 2.00 (sub-industry-neutral)** after controlling for momentum, value, quality, leverage, beta, and industry effects, clearing the conventional |t|>2 significance bar under both neutralizations and confirming the signal prices the cross-section independent of portfolio construction. An equal-weight blend of the factor-neutralized narrative and factor-neutralized momentum target weights -- presented to illustrate the complementarity of the narrative signal with a standard style factor -- cuts the comparison momentum portfolio's drawdown by more than half while raising the information ratio above factor-neutralized momentum on its own (1.10x industry-neutral; 0.80x sub-industry). The factor is robust to activation parameter variation, delivering a positive factor-neutralized IR across all 9 combinations of proximity threshold (40-60% on the 30-day discussion-level gate; 75-85% on the 90-day discussion-level gate) and trailing-high lookback window (120-240 days) on either gate, under both neutralization specifications.
As discussed in Section 9, results in this paper reflect a May 6, 2026 transition of Hivemind's underlying exposure engine from v1 to v2 (which removes potential sources of leakage at some cost to backtest performance) and a May 20, 2026 migration of the activation rule from a single 95%/180-day proximity gate to the two-gate (90-day at 80%, 30-day at 50%) AND-combined rule used.
Keywords: Asset Pricing, Factor Investing, Narrative Factor, Narrative Economics, Thematic Investing, Alternative Data, Natural Language Processing, Alpha Generation, Momentum, Market Neutral, Cross-Sectional Returns