Slow loss recognition inflates a FIDC share's Sharpe ratio against the CDI. Espindola gets that right. He has yet to show that investors earn an illiquidity premium, and his proposed regression would have trouble establishing one. The paper estimates nothing, an omission it acknowledges by presenting the empirical agenda as the research's next step. Its measurement critique deserves attention. The test design deserves scepticism.
Who gets paid, and who takes losses?
FIDCs buy Brazilian receivables, including trade bills (duplicatas), consumer loans and supply-chain credit. These credit rights must account for at least 50% of net assets. The funds issue senior and subordinated classes, with a mezzanine class increasingly placed between them. Cash goes first to the senior class until it reaches a CDI-linked target, then to mezzanine. The subordinated class receives what remains. Losses travel in reverse. Often the cedente, the firm selling the receivables, holds the subordinated class itself. The paper invokes DeMarzo to read that holding as a credible quality signal, and Holmström-Tirole to read it as a commitment to monitor.
Retail participation has grown fast in a large market. According to the ANBIMA figures the paper cites, net assets reached R$ 741.1 billion in the twelve months to November 2025, a rise of 22.5%. Retail accounts increased from 2.4 thousand to 34.3 thousand over the same twelve months. These investors became eligible under CVM Resolution 175, which took effect on 2 October 2023.
Espindola makes two linked claims. Drawing on the corporate-bond decompositions of Elton et al. and Longstaff, Mithal and Neis, he divides the spread over CDI into expected credit loss, a systematic credit premium and an illiquidity premium. He then argues that conventional performance measures overstate the risk-adjusted return. In his account, a FIDC share accrues the receivables rate each day, rising almost linearly until recognition of portfolio deterioration causes a discrete drop. This depresses measured volatility and lifts the Sharpe ratio. Scaling a monthly Sharpe ratio by the square root of twelve compounds the inflation (Lo). Using CDI as the benchmark also attributes all compensation for credit risk and illiquidity to the manager.
His proposed remedy has three steps, using CVM's monthly fund reports. The first assembles a panel by share class. The second estimates each class's smoothing with the Getmansky-Lo-Makarov model, which represents observed returns as a moving average of true returns. The third regresses realized excess return over CDI on term and credit factors using Dimson lagged betas. It then divides the intercept by redemption term, lock-up and closed-class status. The agenda yields four hypotheses.
A promised spread differs from a realized return
The decomposition slips between two quantities, with consequences for step three. In the body, Espindola decomposes the promised spread. This is correct: a senior target of CDI plus a margin must cover expected loss before providing compensation. In the conclusion, he applies the same sum to the share's excess return. For the step-three regression, that means realized return over CDI.
That second formulation is wrong.
Realized returns have already absorbed realized losses. Expected loss is a cost; it earns nothing. With realized excess return on the left and term and credit factors on the right, the regression intercept should contain the illiquidity premium alongside any manager skill and noise. Expected loss has no place in that intercept.
The paper describes at least three components without naming a tax term, although its cited sources give reason to include one. Elton et al. attribute some corporate spread to taxes. The paper also says Law 14.754 of December 2023 changed FIDC taxation and altered the relative appeal of structures. Comparing spreads pre-tax and post-tax across that date mixes tax regimes.
The tail beneath a smooth NAV
Espindola is strongest when connecting smoothing with tranche risk. Citing Coval, Jurek and Stafford, he explains that losses reach a senior class only when many debtors default together. The payoff resembles selling crash protection.
One simulation he cites offers reassurance that depends on the very parameter Coval, Jurek and Stafford warn about. Pinheiro and Savoia simulated structures with 15% subordination across five credit types. The senior class met its target with 99% confidence in every scenario. Such results depend on the assumed default correlation. Coval, Jurek and Stafford show how small errors in that parameter can generate large errors in tranche risk. Espindola also observes that concentrated receivables books produce discontinuous losses: a single credit event at a large cedente or sacado can exhaust the subordination.
Discontinuous losses also complicate step two. A NAV that accrues steadily before jumping when a provision is booked differs from the moving-average process assumed by Getmansky-Lo-Makarov. Fitting that process may leave tail risk understated even after unsmoothing.
Lock-up shares its coefficient with credit risk
Under hypothesis one, senior excess return increases with redemption term and closed-class status, controlling for delinquency and subordination. Espindola acknowledges an identification problem for non-bank lenders and extends it to FIDCs: borrowers paying the highest spreads are also the hardest to resell. The premia are therefore correlated and require careful identification. Beyond delinquency and subordination controls and an intercept split by contract terms, we did not find an identification strategy in the paper.
Hypothesis two exposes the weakness in that split. It predicts greater autocorrelation in less standardized portfolios. These include the non-standard funds created by Instruction 444 of 2006 for overdue, litigated and future receivables. Suppose those assets are held mainly in closed classes with long lock-ups. The redemption-term coefficient would then capture harder valuation together with greater credit risk. Part of the premium found under hypothesis one would be mis-marked credit.
Hypothesis four predicts spread compression following the arrival of retail investors and new managers. Manager numbers increased from 372 to 426 between December 2024 and November 2025. Separating the effects over time presents a similar problem. Resolution 175 took effect in October 2023; the tax law followed in December. Existing funds had until 29 November 2024 to adapt. A dummy covering that period would absorb all three changes.
Espindola discusses survivorship bias in the performance section. Distribution rankings generally include only active funds, while funds shut after credit events disappear from databases. His three-step agenda does not explain how closed classes enter the CVM panel. We also did not find an account of how it handles classes closed before the panel date. The 3,068 FIDCs operating in February 2025 are survivors by construction.
Before paying for the lock-up
We could not test these claims. Our data contains neither FIDC share-class NAVs nor the CVM report fields for subordination and delinquency. It also lacks a CDI series. Substituting US credit ETFs would strip out the subordination and accrual smoothing at the centre of the paper.
An allocator needs realized loss rates by class and vintage, including dead funds, compared with promised spreads. The next comparison should pair senior classes holding the same type of receivables and differing only in lock-up. Without those pairs, higher returns in closed classes are equally consistent with weaker collateral.
The paper belongs on a due-diligence desk for one reason: it explains why a 34.3-thousand-account retail base may be treating a short put on Brazilian credit as a bank deposit.