Shorting 30-day Argentine dollar futures looked tempting between September 2019 and April 2025: contract prices exceeded the spot rate subsequently used for comparison on 91% to 96% of trading days. Siri and Dapena's study offers no more tradeable-looking figure. Turning it into a P&L is where the paper has least to offer.

Measuring the gap

DLR is the A3 Mercados US dollar future, a monthly contract for USD 1,000. Settlement is in pesos against the central bank's Communication A 3500 reference rate. Covered interest parity sets the futures price at spot multiplied by one plus the peso rate, divided by one plus the dollar rate, over the contract's horizon. Carry therefore accounts for much of the quote. The authors ask what information about spot remains beyond it.

Their constant-maturity 30-day series uses linear interpolation between settlement prices for the front two contracts. It contains 5,741 daily observations, spanning January 2003 to September 2026. The spot series uses A 3500 until 2015, then the wholesale close from 2016. The authors divide the sample into six regimes: I is a managed float; II is the first cepo, the Argentine term for exchange controls; III has no controls. IV covers the second cepo, V combines controls with a crawling peg, and VI covers the current bands.

The analysis uses three measures. First comes the realized spread, ln(F_t/S_{t+30}): a positive value means futures priced devaluation above what subsequently occurred. The authors then regress realized log devaluation on implied log devaluation. They test a slope of one with Newey-West errors and 21 lags. Finally, Theil's U compares the futures RMSE with the RMSE of a no-change forecast. The slope test rejects in five of six regimes. Mean spreads show underprediction in II (-0.7%) and III (-0.5%), followed by overprediction of 1.2% a month in both IV and V. The paper concludes that DLR "measures expected devaluation adjusted for risk and interest rates, not an unbiased forecast."

How can regime IV beat a random walk while overshooting by 1.2% a month?

At 0.39, regime IV has the paper's lowest Theil's U. Yet its 1.2% monthly overshoot matches V for the largest, with 6.3% implied devaluation against 5.1% realized. The weak benchmark allows both results. Realized devaluation in IV averaged 5.1% a month, leaving a no-change forecast to miss by roughly that amount on average. Futures carrying a reported 49% implied annual rate can beat that forecast and still miss by 1.2 points. The 49% figure does not reconcile with the 6.3% monthly figure; we return to that below.

In the authors' account, futures "captured the trend of a spot rate that was being adjusted gradually." Carry alone can generate that pattern. We did not find a random-walk-with-drift or interest-differential benchmark. Either would provide the comparison needed to establish whether DLR contributes anything beyond the rate curve.

The floating regimes give a different picture. In regime I, U is 1.02 across 2,181 days, putting futures slightly behind no-change. Section 6.4 nevertheless groups I with III and VI as predicting only marginally better than the current spot rate.

The unreported intercept holds the 1.2% overshoot

The authors draw attention to the pooled slope themselves: 1.06 (t = 0.70). Their abstract says this result "conceals strong regime heterogeneity". Section 6.4 also notes the influence of large devaluation episodes. Both qualifications are justified. Slopes across regimes range from 0.25 to 3.40. The 3.40 in II coincides with the January 2014 and December 2015 jumps; December 2015 also marks the boundary between II and III. We did not find a version excluding jump months or using non-overlapping windows.

The intercept presents the sharper issue. IV has a slope close to one, 1.16 (t = 2.45, still rejected), despite its 1.2% monthly overshoot. Alpha is where that gap belongs. We did not find estimates of alpha, or a joint test of alpha = 0 and slope = 1, anywhere. That joint restriction is the standard unbiasedness test. Testing the slope alone can accept a forecast with a constant monthly error.

VI illustrates the point. Its slope of 0.90, t = -0.19, is not rejected. Even so, the mean spread is +0.6% a month, and futures exceeded realized spot on 62% of days. The full sample also passes (1.06, t = 0.70), alongside a +0.2% mean spread.

VI supplies the only regime-level non-rejection, based on 336 days or roughly 16 independent 30-day windows. On that sample, a t of -0.19 tells us more about power than efficiency.

Buyers paid for cover. Sellers' costs remain unmeasured.

For importers hedging in IV or V, the ex-post cost of cover was 1.2% a month. Applied to the paper's USD 100,000 example, that comes to about USD 1,200 monthly. The authors describe a premium "collected by those who sell cover". They connect it to the worked example's lesson that hedging removes risk, not ex-post gains or losses.

The figures go no further on the seller's side. The 91% and 96% observations come from overlapping daily windows, rather than independent bets. IV's average spread was 1.2%, with dispersion of 4.7%. Its dates extend through December 2023. In that month, the realized spread touched the chart's -50% floor, while the 21-day average realized devaluation reached 77%, against 61% implied. If those days fall in IV, as the dates suggest, its 1.2% mean includes that loss in-sample. The paper deducts no transaction costs, margin funding or roll. And the authors acknowledge that their regime classification is exogenous.

A cepo begins in public. It ends with the jump.

Basis adds another problem. DLR settles against A 3500, whereas the paper uses the wholesale close for spot after 2015. The size of that gap is not reported.

We could not test any of this ourselves. We lack the DLR settlement history and both the A 3500 and wholesale peso series. Substituting other currency futures would remove the Argentine controls and rate differentials behind the effect.

What explains the premium in 2019-25?

The paper expressly limits its claim about the source. Its limitations section says the analysis "documents the sign and size of the bias in each regime but does not identify its source." The candidates it names are "a risk premium, convertibility risk or expectations about regime change." Yet Section 6.3 says that in 2019-25 "the futures price reflected the peso interest rate plus a risk premium". The conclusion's premium "collected by those who sell cover" reads as ex-post accounting. The limitations statement is the defensible claim.

Section 4.2 supplies the authors' own alternative explanation. Under administered regimes, they write, futures reflect the peso rate plus the market's assessment of a regime change's probability and size. Events inside the 30-day window need not match that assessment. Repeatedly pricing a devaluation that mostly fails to arrive within 30 days fits this mechanism just as well as a risk premium. December 2023 brought the jump. Likewise, under floating, a U near one establishes only that this forecast missed the jumps. The paper attributes that result to jumps that "cannot be anticipated."

Several figures still need reconciliation. Our arithmetic puts IV's 6.3% monthly implied devaluation at roughly 76% annualized, compared with the reported mean implied rate of 49%. V's 3.3% and 40% agree, as do the figures for I, II and VI. III is less close, at about 32% against 26%. The full-sample row has the same problem: 2.3% a month against 19%.

A non-overlapping, post-cost version of IV and V would change my view, provided it included a joint intercept test and funded margin at the peso rate. If the 1.2% survives those conditions, sellers of cover earned a premium.