A long-green, short-brown energy ETF basket tended to lose when US climate policy uncertainty rose across the 131 months from May 2015 to April 2026. Kunjal puts the green-minus-brown portfolio's coefficient at -4.2941 on the monthly percentage change in uncertainty. In single-regime OLS, its Newey-West SE is 1.4346 and p = 0.003. The uncertainty measure is the US Climate Policy Uncertainty (CPU) index, a count of newspaper articles combining climate-policy and uncertainty terms. A one-standard-deviation CPU change (0.4060) translates into roughly 1.7 percentage points of brown outperformance that month.
This is contemporaneous exposure. The paper makes no forecasting claim, and the author explicitly says the result does not imply that investors receive compensation for bearing the uncertainty.
Kunjal examines 10 US-listed energy ETFs over those 131 months. He asks whether their returns respond to climate policy uncertainty, whether the response changes between calm and turbulent volatility states, and whether investor sentiment magnifies or dampens it. The behavioural premise comes from the firm-value and IPO studies he cites (Bagh et al., 2025; Amini et al., 2025). In those studies, pessimistic sentiment amplifies climate risk's effect.
His sentiment measure is the American Association of Individual Investors (AAII) weekly bull-bear spread, averaged to monthly: the bullish share of respondents minus the bearish share.
Brown funds gain as uncertainty rises
The sample contains five green ETFs (ICLN, QCLN, PBW, PBD, SMOG) and five brown ETFs (FCG, IXC, XOP, XLE, VDE). The regressions explain each fund's monthly log return with six variables:
- its own lagged return;
- the change in the Gavriilidis et al. CPU index;
- the change in the AAII bull-bear spread, plus its product with the CPU change;
- the S&P 500 return and the Brent return.
A two-state Markov switching model allows the intercept, CPU slope and residual variance to vary by state. It has 13 parameters. According to the author, the fully switching 18-parameter model produced multiple local maxima.
The mechanism deserves attention. The CPU index rises during debate about loosening or reversing climate policy as well as tightening it. Higher uncertainty can therefore signal lower expected compliance costs for oil and gas producers. Kunjal also proposes a real-options channel: uncertainty postpones low-carbon investment and prolongs the life of existing fossil assets. Both channels predict a non-negative response from brown funds, contrary to the stranded-asset prior.
The estimates fit that account. Single-regime CPU slopes for brown funds range from 2.4635 (IXC) to 5.5832 (FCG); every Benjamini-Hochberg q is below 0.001. Green slopes span -1.4729 to 2.2047, with q no lower than 0.241. For the equal-weighted green portfolio, the coefficient is 0.3851 (p = 0.724). The funds ostensibly most exposed to climate policy have no measurable exposure to uncertainty about it.
The author flags an important qualification early. Brown funds' return correlations average 0.95, reaching 1.00 for XLE and VDE. The five brown findings are thus close to a single finding. The portfolio test addresses that dependence, and the result survives in the single-regime model and calm state.
What do the election dummies leave standing?
The abstract acknowledges the difficulty. With dummies for November 2016 and November 2024, the low-volatility brown coefficient remains significant at 5% only for IXC (3.8178). VDE reaches 10% (2.0443). FCG drops from 3.8564 to 2.4137, and XOP from 4.3768 to 2.4456. XLE declines from 2.0697 to 1.8620. All three lose significance.
Kunjal's conclusion interprets this concentration as evidence for the deregulation channel. In the same sentence, he says it does not confirm the full-sample finding. With the dummies included, high-volatility coefficients for FCG (7.7201) and XOP (6.7411) remain significant at 5%. IXC moves from 10% (2.1689) to 5% (2.6830).
The signs all survive. The dummies cost precision, with every point estimate still positive. Even so, two of 131 observations account for a large share of the calm-state slope for FCG and XOP.
A desk would want the headline portfolio tested with those dummies. We did not find either the single-regime or green-minus-brown regressions re-run on that basis. The election check covers the ten individual funds in the switching model; the portfolio table appears to retain the full sample. If deregulation explains the result, election clustering is what one would expect. A monitoring rule should then anticipate exposure arriving in event months.
An energy rally inside the turbulent state
For brown funds, the high-volatility state begins between October 2019 and February 2020 and finishes between September 2022 and November 2023. It includes the COVID crash and the 2021-2022 energy run. VDE's average monthly return is 3.05% in that state, versus -0.38% in the calm state. These figures support the author's choice to label states by residual variance and avoid bull and bear terminology.
The regime split offers little for allocation. Across every brown fund, Wald tests of equal CPU slopes between states produce p-values from 0.165 to 0.537. AIC selects two states for all ten funds. BIC selects a single regime for ICLN, PBW, SMOG, IXC and XOP. The text describes brown high-volatility point estimates as larger, though IXC runs the other way: 2.1689 versus 3.9311 in the calm state. Green-minus-brown has a calm-state slope of -5.0979 (SE 1.3335). In the high-volatility state it is -3.5395 (SE 3.0697), an insignificant estimate.
Sentiment supplies no conditioning variable either. For all ten funds, the CPU-sentiment interaction is insignificant at 5%. The change-in-spread measure has kurtosis of 18.99 and behaves poorly because the bull-bear spread crosses zero and changes sign. The author acknowledges the extreme values.
He separately re-runs the model using the mean-centred level of the spread, since the sentiment hypotheses concern optimism versus pessimism. That version also leaves the interaction insignificant at 5% for all ten funds. Only PBD, FCG and XOP register 10% hits. We accept the null at face value. AAII captures equity mood, and Kunjal suggests that a climate-specific measure might pick up an effect this measure misses.
Before election months, check the brown CPU slope
The surviving finding concerns holdings. Brown funds have Brent betas from 0.4008 to 0.6867, alongside a positive CPU slope. Green funds have market betas from 1.0791 to 1.4464 and no CPU slope. Combining XLE with VDE adds no diversification. A transition tilt that buys green and shorts brown implicitly shorts CPU spikes. At mean sentiment, a one-standard-deviation CPU change moves brown funds by 0.85 to 1.64 points a month in the calm state. The author also discloses that selection used AUM as of July 2026, excluding funds that closed during the sample.
We could not test the result on our data. We lack both the CPU index and the AAII survey; constructing a substitute from news would measure something different. The paper also ends in April 2026, beyond our history's 2025 endpoint.
Until the green-minus-brown portfolio is re-run with the November 2016 and November 2024 dummies, treat the calm-state brown CPU slope as an election-month exposure.