In their paper titled Complicated Firms, Lauren Cohen and Dong Lou showed how the flow of information such as an industry shock first affects the stock prices of firms that generate all revenue from that industry, followed later by the stock prices of conglomerate firms that generate only a portion of revenue from it. They documented that because of the delay in information processing—it’s challenging for investors to discern the industry-specific revenue source within conglomerates—there is a predictive relationship between the returns of companies operating in a single industry (pure plays) and those of conglomerates with exposure to the same industries.
We originally applied Cohen’s and Lou's framework in our 2020 analysis (The Challenge of Calculating Industry Momentum for Diversified Firms) and found evidence consistent with their hypothesis that industry-level trends are slower to be reflected in conglomerate firm prices. Now, given several years of turbulent market conditions since then, we again test whether the signal remains valid and explore methods of enhancing the factor with FactSet’s RBICS with Revenue framework. For illustrative purposes, we selected conglomerate firm Siemens and highlighted in the following table the RBICS with Revenue exposure breakdown for its 2025 fiscal year.
Baseline Methodology
We first constructed a baseline daily factor using a methodology similar to our previous analysis. We grouped the constituents of the Russell 3000 into pure-play and conglomerate firms. Pure-play firms are companies that derived at least 75% of their total revenue from a single RBICS L4 industry. We then created RBICS L4 return benchmarks, calculated as the equal-weighted average of the previous day's total returns for all pure-play companies classified in that L4.
The L4 return benchmarks were then used to compute a synthetic return for each conglomerate company. It is the sum of the L4 benchmark returns the firm generates segment revenue from, weighted by the firm’s L4 revenue exposures where rev_exp(s) is the RBICS L4 revenue sector exposure.
Conglomerates are ranked based on their synthetic returns and are then assigned to quintiles. Equal-weighted portfolios are formed for each quintile, with forward returns measured over holding periods of 1, 2, 5, and 21 trading days. L4 benchmarks with fewer than three pure plays are assigned the universe mean return.
Signal Enhancement Testing
Four enhancements to the baseline factor were also tested.
Enhancement A (Excess Return) subtracts the daily market-cap-weighted universe return from each sector return before computing the company signal, isolating sector-specific returns from the broader market.
Enhancement B (Market-Share Weighted) weights each pure-play stock by its RBICS L4 revenue amount when calculating the L4 benchmarks.
Enhancement C (Z-Score) divides each company's aggregated signal by its rolling 21-day standard deviation, converting the raw signal to a z-score and removing the bias toward companies exposed to high-volatility sectors.
The combined A+C (Sector-Level) variant applies the volatility normalization at the sector level—specifically, to the excess sector return before aggregating to the company level—ensuring that input signals are comparable before any aggregation occurs.
U.S. Backtest Results
We ran a backtest against the Russell 3000 universe from January 2016 to December 2025. We also constructed a long-short portfolio by longing the highest-ranked conglomerate quintile (Q5) and shorting the lowest-ranked quintile (Q1). The chart below presents the long-short Sharpe ratios for all five variants with holding periods of 1, 2, 5, and 21 trading days.
Annualized L/S Sharpe Ratios by Variant and Holding Period—Russell 3000 (1-Day Signal Horizon, 2016–2025)
The baseline factor produced the strongest 1D Sharpe of 1.032. Enhancement C, which applies rolling z-score normalization to the company-level signal, produced a comparable 1D Sharpe of 0.989 and had the strongest performance with the 2-day horizon, where its Sharpe of 0.632 exceeded the baseline's 0.505. The excess-return construction (Enhancement A) achieves 0.943 at the one-day horizon, while the A+C sector-level combination produces 0.896.
Below are the cumulative 1D quintile returns over the testing period for the baseline variant.
These results support our core hypothesis: at the 1D holding period, the highest-ranked conglomerates quintile (Q5) consistently outperforms the lowest-ranked quintile (Q1).
Despite that, the signal deteriorates sharply at longer holding periods across all variants. In the US, figures suggest the predictive content of the industry momentum signal for conglomerates is restricted to a short window following the pure-play peer return observation.
The rapid decay calls for a closer look at the signal's construction. Using one-day sector returns captures only the most recent price movement of pure-play peers; it’s a noisy, short-lived signal.
Cohen and Lou relied on one-month cumulative returns in their original work, reflecting that information spreads gradually across firms over time. Additionally, the universe is US-domiciled companies only and is populated by many thinly traded small-cap stocks.
Together, the results suggest that testing this factor against a global large and mid-cap universe over a longer signal horizon is likely to yield more meaningful results.
Universe and Signal Design
Expanding to a global universe offers several advantages for testing the industry momentum factor. It includes developed and emerging markets, leading to a substantially richer set of well-populated L4 benchmarks than is available in the US domestic universe alone. The universe's large and mid-cap composition means the pure-play stocks in the sector benchmarks are predominantly liquid, continuously priced securities.
The graph below shows the pure-play vs conglomerate firm count for the global benchmark universe.
The pure-play return signal horizon extends to 21 trading days, which is approximately one calendar month. Instead of using yesterday's 1-day return as the predictive signal, the L4 benchmarks use the equal-weighted sum of the 21-day cumulative return of pure-play stocks. The same four signal enhancements are applied with identical construction logic; only the investment universe and signal look-back window differ.
Global Backtest Results
We ran a backtest against the MSCI ACWI for the same January 2016 to December 2025 range. The next table presents the long-short Sharpe ratios for all five variants with the 21-day signal horizon. The improvement relative to Russell 3000 results is substantial and consistent throughout variants and holding periods.
Annualized L/S Sharpe Ratios by Variant and Holding Period — MSCI ACWI (21-Day Signal Horizon, 2016–2025)
The baseline Sharpe ratios are materially higher at every holding period, but the improvement at longer holding periods is particularly noteworthy. The 5-day Sharpe in the global universe increased to 0.768 from 0.207 against the US-only universe, and the 21-day Sharpe increased to 0.344 from 0.105. Against the global large and mid-cap universe, our industry momentum signal demonstrates meaningful predictive power over multiple weeks.
The A+C sector-level enhancement led at the 5-day horizon with a Sharpe of 0.906, while Enhancement A (excess sector return) edged slightly ahead at the 21-day horizon with a Sharpe of 0.552. Enhancement A consistently adds value relative to the baseline across all holding periods, confirming that removing the common global market component from sector returns isolates a more durable source of predictability regardless of rebalancing frequency.
The chart below shows the quintile returns over the testing period for the A+C enhancement with a 5-day rebalance period.
Our analysis revisits the industry momentum factor for conglomerate firms and extends the original framework through a series of signal enhancements. Against the US universe, the factor demonstrates strong short-term predictive power, but it decays rapidly beyond the 1-day horizon.
Shifting to a global large and mid-cap universe with a 21-day signal horizon produces a substantial and consistent improvement across all variants and holding periods. That indicates the gradual information diffusion from pure plays to conglomerates is better captured in a regionally broad and liquid market setting.
Throughout this analysis, FactSet's RBICS with Revenue dataset proves essential. Its granular six-level taxonomy enables precise identification of pure-play firms and accurate decomposition of conglomerate revenue exposures—distinctions that traditional single line classification systems are incapable of making.
Mapping business segment revenues to a standardized granular classification taxonomy ensures that each conglomerate's synthetic return reflects its true business mix, and the dataset's broad global coverage is what makes the expanded global universe viable. RBICS with Revenue is purpose-built for any analysis that demands quantified multi-industry exposure profiles, making it a versatile and powerful tool for systematic investors.
This blog post is for informational purposes only. The information contained in this blog post is not legal, tax, or investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.