Raising the weighted compliance target from 0.4 to 0.6 takes a collective investment pool's certainty-equivalent loss from 2.72% to 48.65%. For Chen, Laudagé and Zhu, however, the distributional result tells us more about wealth sharing inside the pool than about regulation. The authors themselves assign most of the differences in investors' costs to the Pareto weights.

The abstract closes with the claim that regulatory uncertainty "can also create substantial disparities in how its costs are distributed." A table row gives our answer. Among the four-investor runs, identical γ=3 and β1=0.7 leave every investor with the same 17.41% loss. Even so, d1 is -31.06%. Any proportional cost would produce that disparity: d_i captures the distance between the sharing rule's allocation and capital share. Regulation does not create the gap.

Wealth flattens at the floors

A single manager invests a pool on behalf of several investors. Each has constant relative risk aversion (CRRA) utility. The collective utility takes the highest Pareto-weighted sum of their utilities across all possible wealth allocations. Trading takes place in a complete Black-Scholes market with one stock, one bond and continuous trading. Regulatory uncertainty enters through the terminal wealth floor: the manager does not know which floor will apply. Compliance therefore requires a weighted average of probabilities, α1·P(X ≥ L1) + α2·P(X ≥ L2) ≥ a.

The authors interpret the thresholds as belonging to different business lines. Each line has an implied threshold; premium or provision shares supply the weights.

Using the martingale method, they choose terminal wealth as a decreasing function of the state-price density. The two-threshold payoff has a distinctive shape. Wealth follows a Merton-type branch in the best states, using a budget multiplier different from the unconstrained one. It then flattens at L2 over one band and at L1 over a lower band. In the worst states, wealth follows another Merton branch below both floors.

This tail behaviour is the Basak and Shapiro result cited in the paper. VaR-style protection is purchased for intermediate states, then abandoned in the tail. The multiplier determines whether the middle Merton region or the L1 plateau disappears. Letting α1 go to zero formally recovers the single-threshold VaR solution.

The weighted constraint guarantees probability a at the lower threshold alone. The probability of clearing L2 can be well below a. The paper explicitly says that each threshold need not be met with probability a. Its defence is a regulator who randomly selects a threshold with probabilities α_j, independently of markets. Under that interpretation, the payoff clears the selected threshold with probability at least a. The defence depends on having the right α_j. The authors take those weights as known and leave uncertainty about them for future work.

There is no data. The numerical illustration sets T=1, r=1%, μ=5%, σ=20%, L1=3, L2=5, x0=2.5 and a=0.4. For the single-investor baseline, α1=0.7 and γ=3. Initial wealth is below discounted L2, which rules out buying the higher floor risk-free.

From 2.72% to 48.65% as the target moves from 0.4 to 0.6

The baseline certainty equivalent drops from 2.5423 to 2.4731, costing 2.72%. The authors describe many settings this way: "the payoff shape changes visibly while the welfare loss remains economically moderate." Their abstract and conclusion give prominence to the large-loss cases.

The nonlinear response is the finding that matters. Targets of 0.2, 0.4 and 0.6 produce losses of 0.38%, 2.72% and 48.65%. With a=0.6, the certainty equivalent falls to 1.3054. Raising L1 to 4 costs 33.05%; reducing x0 to 2.0 costs 25.47%. A lower α1 of 0.5 costs 17.42%, compared with 1.21% at α1=0.9. The loss vanishes at L1=2 or x0=3 because the Merton payoff already satisfies the constraint.

At x0=3.0, compared with the baseline 2.5, the Merton payoff likewise complies and the loss is zero. These are parameter changes made individually from the same baseline. A practitioner should read them as sensitivity to misspecification. Moving a from 0.4 to 0.6 increases the loss 18-fold, from 2.72% to 48.65%. Cutting α1 from 0.7 to 0.5 raises it 6.4-fold, to 17.42%. Moving α1 up to 0.9 only brings it down to 1.21%.

The upper threshold barely matters.

Setting L2 to 4, 5 or 6 leaves the certainty equivalent unchanged at 2.4731. The appendix explains this through α2=0.3 and the expense of reaching L2 from x0=2.5. Risk aversion has more effect: the loss rises from 0.73% at γ=1.5 to 4.93% at γ=4.5.

Who pays in the four-investor pool?

The authors use a lower α1 of 0.5 for the four-investor runs. The homogeneous pool consequently loses 17.42%, matching the single-investor result at α1=0.5. In the collective section, rc_i denotes each investor's percentage certainty-equivalent loss. A separate measure, d_i, subtracts her Pareto weight from her share of the total loss. The paper sets that weight equal to her capital share.

Consider identical risk aversions of 3, with β1=0.7 for the large investor. Every investor loses 17.41%. Nevertheless, d1 is -31.06%, while each small investor has d of +10.35%. Equal percentage losses expose what the sharing rule does: it allocates about 39% of terminal wealth to the large investor despite her 70% capital share.

The paper acknowledges the mismatch. Capital-share weights, it states, do not equate allocations' market values with contributions. Under the same rule, comparison with stand-alone investment leaves the large investor 44.38% worse off for participating. Each small investor gains 103.54%. The authors supply a closed form for this equal-γ case. Faced with those weights, the large contributor would decline to join.

Scenario 4 introduces different preferences: γ=1.5 for the large investor and γ=4.5 for the other three. The resulting d1 is -31.25%, just 0.19 points from the identical-preference result. The abstract's statement that small contributors bear disproportionate shares is true as reported. Most of that result comes from the sharing rule. The authors themselves identify unequal Pareto weights as the source of the disparities. With equal weights, the largest d is -5.14%.

Risk aversion does affect how much the pool loses. At equal Pareto weights, the loss is 9.70% when three of four investors have γ=1.5. It reaches 22.54% when three have γ=4.5. In scenario 3 with β1=0.7, the lone γ=4.5 investor's participation gain is 247.30%. That comparison scales thresholds by capital share, a normalisation chosen by the authors.

Daily SPY/SHV replication is still under evaluation

We are constructing a daily strategy using SPY for the risky asset and SHV for the defensive asset. The paper makes no empirical claim. Our construction follows its payoff. SHV has rate and price risk, so it cannot serve as a risk-free bond, and daily rebalancing cannot reproduce continuous replication. We choose the thresholds, weights, risk aversions and Pareto weights as research scenarios. They come from no regulator or beneficiary dataset. Meeting the fitted model's probability constraint offers no guarantee of historical compliance. Evaluation is still underway; we have no results to report.

The paper's formulas identify the obvious stress point. Its plateaus behave like digital payoffs. In the maximum-compliance case, the replicating position contains L2·e^(-rτ)/(σ√τ)·φ(d). Near a cutoff, that term grows without bound as the remaining time shrinks. Daily trading is coarsest precisely there. The constraint applies only at the T=1 horizon, leaving wealth unconstrained before then.

A useful backtest must report horizon threshold hit rates and trading costs near the cutoffs alongside returns. A realized weighted compliance rate meeting or exceeding the 0.4 target after costs would change our view, provided the horizons did not all end in rising markets. The paper's own table also cautions against demanding more compliance: moving from 0.4 to 0.6 raised the single-investor loss from 2.72% to 48.65%.