The Hidden Costs of FX Exposure : A Performance Attribution Analysis of the Errors in FX Impact Reporting

Sammanfattning

The accuracy of communicated foreign exchange (FX) impact figures matters because theyflow directly into the headline measures that investors and executives rely on most. Multinational industrial groups commonly report three types of FX impact: the transactional impact on the subsidiary’s income statement, the translation impact recognised in other comprehensive income, and the constant-currency impact presented as an analytical overlayon the consolidated income statement. In practice, these figures are constructed by translating income statement items at either daily closing rates (Method 1) or period-averagerates (Method 2). However, the aggregation conventions on which both methods rely havenot been systematically evaluated against a theoretically exact benchmark.This thesis quanitifies the systematic error introduced by industry FX methods, identifying the conditions under which the choice of method materially affects the reported figures. This is done in collaboration with Sandvik, a Swedish multinational industrial group,whose transaction exposure disclosures and reported financial figures are used to calibratea simulated subsidiary. The Performance Attribution Model (PAM) developed by Blomvalland Hagenbjörk (2019) is implemented as the benchmark, representing the subsidiary’soutstanding monetary items as zero-coupon bonds and producing a residual-free FX attribution that captures exchange-rate movements continuously, including the pre-recognitionexposure windowbetween ordercommitmentand accountingrecognition,heretermedthebill-of-materials (BOM) window which the industry methods cannot observe.This was done by usage of a Monte Carlo simulation with K = 5,000 iterations, conductedover a 2007–2025 window, generating independent transaction datasets processed by allmethod variants and by PAM in parallel. The simulation was calibrated year by year toSandvik’s reported revenue growth, gross margin, and transaction-exposure currency mix.Four variants of the transactional and translation impacts were evaluated alongside threevariants of the constant-currency impact, using meanerror, rootmeansquareerror(RMSE),confidence intervals, and kernel density estimation.The results differ by impact type. For the transactional impact, all variants are close toPAM onaverage, but RMSE rises more than fivefold from daily to quarterly averaging, sothe quarterly-average convention should be avoided in favour of daily, weekly, or monthlyrates. For the translation impact the industry methods consistently overshoot PAM acrossthe period, while the averaging window barely matters, RMSE changes by only around4% across variants, showing a systematic overstatement rather than an aggregation effect.For the constant-currency impact the reference rate matters more than the window, theaverage-rate variant performs almost identically to Method1, whiletheclosing-ratevariantadds noise without reducing bias and should be avoided.The largest error, however, is structural. The industry methods only track FX exposurefrom the point of accounting recognition, but the real exposure begins at order commitment, an order-to-delivery window that the methods cannot see. Benchmarking againstPAMbonds+BOM, which captures this pre-recognition exposure, reveals a large and systematic error, one that lies not in how the methods aggregate, but in what they never measure.