Selling equities when regulatory news spikes could mean selling just when expected returns are highest, if regulatory uncertainty earns a premium. That tension runs through this paper, though the authors never frame it this way. Bäuerle, Branger, Chen, Gehde-Trapp, Mahayni, Schienle and Sureth-Sloane offer a research agenda without an empirical claim. Their disclosure is explicit: "No empirical datasets were generated or analyzed during the current study." They report no Sharpe ratio, sample period or estimated coefficient. All the backtest figures below belong to a strategy we constructed ourselves; none belongs to the authors.
The paper sets out the programme for DFG Research Unit 5583 (FOR 5583), "Asset Allocation and Asset Pricing under Regulatory Uncertainty." Regulation changes payoffs and the portfolios investors may hold. Uncertainty over those rules therefore changes optimal portfolios. It can command a premium only if it covaries with marginal utility or affects whole sectors.
The applications span three domains: sector-specific regulation, principally financial regulation, climate regulation and taxation. Insurance appears throughout. No data are supplied. The authors plan to use newspaper corpora covering CO2-related climate policy, dividend futures and options alongside their underlyings, and a panel of European firms. They specify neither a sample period nor a universe size.
What do the seven authors propose?
Regulatory uncertainty, in the paper's definition, concerns the design, timing, implementation, interpretation and enforcement of rules governing payoffs and permissible actions. It includes capital ratios and solvency thresholds, carbon-pricing rules, loss-offset rules and audit intensity. Broad economic policy uncertainty indices, the authors argue, mix this object with elections and macro policy. They want measures tied to the institutional target of regulation. Table 1 divides the subject into 4 dimensions: future interventions, parameters and scope, interpretation and enforcement, and cross-domain interactions. Climate-related capital effects on banks illustrate the last category. The authors describe the table as non-exhaustive.
They then identify 4 ways regulation enters a financial model. Rules can alter cash-flow dynamics or change beliefs as agents infer a hidden regime from signals. A binding Value-at-Risk or tax constraint can change the feasible set. Regulation can also restrict intermediary capital and liquidity.
These channels lead to the asset-pricing argument. Uncertainty enters prices when it changes expected cash flows, constraints or discount rates. Earning a premium requires more. "It commands a premium only under stronger conditions: when regulatory shocks covary with marginal utility," the authors write. They also describe shocks that affect aggregate investment opportunities or intermediary capital and liquidity capacity. Constraints may tighten in bad states. Sector-wide shocks may be impossible to diversify away.
The remainder lays out 5 planned projects within FOR 5583. These address portfolio choice under ambiguous VaR thresholds, estimation of network spillovers, regime-switching models, a news-based climate-policy uncertainty index, and a measure of tax uncertainty inferred from derivatives. The writing mostly looks forward: "we will", "we aim". Although the availability statement says "All results are derived analytically," we did not find a proposition or derivation in the text.
Direction versus uncertainty in a policy headline
The climate project contains the idea most useful to a trader: separate expected policy, a first moment, from uncertainty about policy, a second moment. An expectation that carbon prices will rise over the next decade reprices brown and green cash flows. Uncertainty concerns when a rule arrives, whether a subsidy survives a court challenge, and which sectors fall within its scope. The authors argue that this second moment affects risk premia and increases hedging demand. It also raises the real option value of delaying irreversible investment.
The cited news indices, Baker et al. 2016 and Gavriilidis 2021, use dictionary-based counts. Both a firm carbon-price path and a stalled bill can generate a counted story, despite carrying different economic signals. The proposed index promises 3 additions: separating direction from uncertainty, classifying policy instruments, and combining newspapers with market data. Supervised classifiers and transformer embeddings would do the work. We did not find a design for validating the labels.
The authors do intend to benchmark their measures "against observable developments in emissions markets and policy events," though this does not provide held-out label validation. The distinction matters. Separating direction from uncertainty is the first of the three promised additions, and its success depends on how the training texts are tagged.
Carbon prices are also rejected as a proxy. The authors say emissions-trading prices reflect market design and risk-bearing capacity, alongside regulatory constraints and market structure. We encountered a related empirical problem in an earlier note on brown energy ETF exposure to policy uncertainty resting on election months. The measured sensitivity there was concentrated in a narrow set of months.
Four channels, four different trades
The four entry points imply different portfolio responses. Faced with an uncertain, tighter capital rule, banks shed risk: that channel favours selling risk. Climate policy's cash-flow channel is relative. In the paper's example, a stricter regime cuts brown cash flows, increases demand for green hedges and produces a regime-risk premium. A climate regime shift therefore suggests a long-short trade without a net equity direction. Through the belief channel, returns depend on filtered regime probabilities and revisions to transition probabilities around announcements. If the conditions for a premium hold, greater uncertainty should also predict higher subsequent returns for exposed assets.
A rule that sells SPY as regulatory uncertainty increases chooses a direction the framework leaves open. The rule assumes the constraint channel wins and the premium is absent or slow to arrive. The paper offers no grounds for that choice.
The authors acknowledge the open questions.
Their conclusion calls for empirical measures that "more sharply separate expected regulatory changes from uncertainty surrounding their timing, scope, implementation, and enforcement," and for work to "test more directly when regulatory uncertainty is priced rather than merely incorporated into expected cash flows," in their words. It also defends regulators' room to respond: uncertainty's costs "must be evaluated against the benefits of regulatory flexibility and adaptation."
The tax gauge leaves its own non-tax wedges unexamined
The fifth project gives the most concrete proposal. Dividend futures pay gross dividends; a replication using puts, calls and the underlying carries investor taxes. Comparing them produces a dividend tax penalty, dtp = 1 − (1 − τD)/(1 − τC). Equal dividend and capital-gains tax rates produce zero. The authors would measure tax uncertainty through time-series variation in dtp, then use it in Fama-MacBeth tests and a panel of European firms' payout and investment. They cite Brav et al. (2005), where roughly one-third of executives describe taxes as important for payout.
The authors criticise bond-based tax measures because liquidity, default risk and embedded calls contaminate them. We did not find comparable scrutiny of the proposed spread. Dividend futures have liquidity effects and a risk premium of their own. Replicating with options introduces funding and borrow costs, early exercise of American contracts and breaks in put-call parity. Each can change dtp without a change in expected taxes. Using variation in a noisy level to measure uncertainty creates another problem: the noise itself registers as uncertainty. The paper does not specify an estimation window.
Our SPY/SHY switch, 2020 to mid-2024
We built a daily SPY/SHY switch for 2020-01-01 to 2024-07-01. With no performance reported in the paper, there is no authors' result to compare against ours. At each session's 16:00 New York cutoff, a frozen sentence-level classifier counts articles connecting a regulatory object with uncertainty. It excludes generic fear and settled policy direction. We divide that count by all eligible articles and average the ratio over 20 sessions.
The portfolio switches to SHY when the score rises above the 80th percentile of its expanding history. It switches back to SPY below the median. Orders execute at the next open. We charge $0.0040 per share in commissions, subject to a $1.00 minimum; all reported figures are net of those charges.
Two design choices matter as much as the signal could. The active ETF position is limited to 50%, and the rest stays in cash earning nothing. Annualised volatility was 12.07%, consistent with keeping half the capital in cash. The threshold also requires 252 prior valid scores beyond the 20-session window. As a result, no switch could occur until roughly early 2021. The portfolio took the March 2020 crash at half weight in SPY while the signal remained silent. Maximum drawdown was 18.67%.
The live switching period thus covers about three and a half years, including one bear market in 2022. Total return was 39.67%, with a Sharpe of 0.73. Those figures describe one automated pass at our own construction and say little about the authors' framework. The run cannot distinguish the constraint argument supporting the exit rule from the premium argument against it. The relevant test is whether returns following high-score days are higher or lower.
The sharpest claim awaits a test
The framework supplies a careful vocabulary. Its sharpest testable claim is that expected regulatory direction and uncertainty about regulation should be priced differently; most of the remaining work is still planned. We would be persuaded by a climate index that separates directional and ambiguity signals on held-out labels, followed by evidence that only the second earns a cross-sectional premium. Until then, a trade's direction on regulatory news remains the researcher's guess.
Our backtest stops at 2024-07-01, and everything after that date is deliberately left untouched so the same strategy can be checked out of sample later.