A brand-tracker update deserves less weight when the survey series keeps jumping around. Raithel backs that rule with two estimates: a -0.086 interaction across 513 firm-years and a -0.051 cumulative response over four quarters across 12 automakers. Both can inform how you read the next update. Neither comes with a portfolio test.

From Smith and Winkler to YouGov

Smith and Winkler's optimizer's curse supplies the mechanism. Pick the alternative with the highest noisy estimate and you systematically select estimates that were too high. The rational response is to shrink each estimate by 1/(1+σ²). Raithel proposes that investors do the same with nonfinancial signals: a rise in brand perception should move the share price less if the brand's survey history is volatile. In Raithel's account, this is efficient pricing. The market anticipates disappointment bias, and the paper claims no excess return. Taking every survey uptick at face value would mean overpaying for noise.

Study 1 draws on YouGov BrandIndex, which surveys about 5,000 US respondents daily. Its six questions cover quality, value, satisfaction, recommendation, affect, and reputation. Each score subtracts the negative share from the positive share; the measure averages all six. The paper defines temporal volatility as the log standard deviation of 60 monthly changes.

Annual Carhart four-factor buy-and-hold abnormal return is the dependent variable. The regression includes changes in brand perception, cashflow, profitability and analyst recommendations, together with each signal's volatility and each change-times-volatility interaction. Controls and firm and year fixed effects enter too. Every variable is standardized. Because the survey starts in December 2012 and volatility requires five years of history, the usable window is only 2017 to 2019. It covers 179 brand-name stocks in 27 industries.

For Study 2, the author uses YouGov's customer satisfaction question and narrows the universe to 12 US-listed automakers. Quarterly data from Q1 2013 to Q4 2021 yield 348 usable firm-quarters. These enter a panel VAR estimated by GMM.

The brand interaction in Study 1 is -0.086 (SE 0.041, p = 0.036); the brand main effect is 0.135. Cashflow's interaction is -0.083, profitability's is -0.096 (p = 0.088), and analyst recommendations' is -0.081. Study 2 reports a cumulative abnormal-return response to the interaction of -0.051 after four quarters (95% CI -0.076 to -0.025).

What does volatility actually discount?

The hypothesis is about the slope on brand changes. In Study 1, volatility alone earns nothing: the brand-volatility coefficient is 0.222 (p = 0.144). The three financial volatilities are also far from significant, with p of 0.480, 0.775, 0.437. High-volatility brands do not underperform in Study 1. Study 2 does show a return drag from volatility alone for two quarters (-0.065 by quarter 2), after which significance disappears. Investors pay less for each unit of brand improvement.

For signal sizing, that distinction matters. The author says the brand effect "becomes insignificant already at mean levels of Vol. ΔBrand Perception." Taken literally, the survey contains information only in the calmer half of the volatility distribution. A practitioner would gate the signal or scale it by inverse volatility. Nobody would short the jumpy names.

The brand interaction holds up under controls

Adding controls and fixed effects does less damage to the brand interaction than to its financial peers. Without either, the brand estimate is -0.075 (SE 0.034); with both, it is -0.086. The financial interactions fall by a quarter to a half: cashflow goes from -0.154 to -0.083, profitability from -0.147 to -0.096, and analyst from -0.108 to -0.081. Pairwise correlations among the volatility measures range from -0.024 to 0.185. That limited overlap argues against the brand term being a disguised financial proxy.

The raw-unit translation helps. On the paper's scale, raw BHAR has a standard deviation of 29.570. By our arithmetic, the 0.135 brand main effect therefore represents about 4.0 points of annual abnormal return per standard deviation of brand change. Each standard deviation of volatility removes about 2.5 points. These associations are in-sample and before costs.

There is little data behind that weight. Three usable years across 179 firms amount to about 2.9 observations per firm, and within R² is 0.166. A control has the model's largest coefficient: competitor-perception volatility, at -0.430 (p = 0.048). The regression is also contemporaneous. It matches calendar-year returns to same-year changes in annual cashflow and profitability, although those changes are published after year-end. We raise that publication-timing objection. Raithel acknowledges a related limit, saying Study 1 "does not account for potential temporal dynamics and feedback loops among the model variables."

Prediction rests on twelve automakers

Study 2 attempts to close that gap by lagging everything one quarter. The interaction Granger-causes abnormal return (χ² = 13.992). Its cumulative response stays negative at every horizon, moving from -0.065 in quarter one to -0.051 in quarter four. For this sample, that answers the dynamics objection.

The sample is 12 firms in a single industry. Study 2 simultaneously changes the measure, frequency, estimator and universe, leaving it unable to isolate what replicated. The paper describes the PVAR as helping to address endogeneity. The result establishes Granger precedence under a Cholesky ordering: it supports prediction while leaving causal identification open. Three details trouble me more:

Data access before alpha

We could not test the result on our own data. Without a historical BrandIndex panel, we cannot construct the 60-month volatility measure. We lack a point-in-time mapping from brands to issuers as well. And without survey release timestamps, a lagged test would require guessing when the information became available.

The clock matters.

News sentiment is the obvious substitute, though it measures press tone. Raithel's mechanism depends on repeated consumer judgments.

Study 1 measures the signal annually across 179 brand-name stocks, despite YouGov collecting it daily. How quickly the signal decays remains open. The Study 1 sample ends in 2019.

I regard inverse-volatility weighting of survey signals as a sound hypothesis, supported by an in-sample interaction of -0.086. One result would change my view: a portfolio sort using lagged, timestamped brand changes where volatility scaling improves the spread net of costs outside 2017 to 2019.