The 6.32% one-day loss threshold reported by Maghfiroh, Aprilianto and Agustin for Bank Syariah Indonesia needs reconciling before anyone uses it to size positions. The paper's daily mean and standard deviation give 5.98% under the normal VaR formula. A simulation drawing from those same inputs should converge to that result.

The sample: PT Bank Syariah Indonesia Tbk

PT Bank Syariah Indonesia Tbk (BRIS), the Sharia-compliant bank created by the 2021 merger, is among the actively traded Sharia-compliant issuers on the Indonesia Stock Exchange (IDX). The study uses 474 daily closes covering 1 November 2023 to 31 October 2025. Returns are simple price returns and exclude dividends. Closing prices ranged from IDR 1,460 to IDR 3,160 and averaged about IDR 2,551. Daily returns ranged from -11.11% to 12.32%; their mean was 0.1465% and their standard deviation 2.6318%.

Excel supplies the risk calculation. The function NORM.INV(RAND(), mean, sd) generates normal returns from the historical mean and standard deviation. The sheet estimates those moments again from the draws, then inserts them into VaR = μ − Zσ at 99%, 95% and 90%. The authors describe 1,000 iterations "per observation" across the 474 observations, with the results averaged. They report one-day VaRs of 6.32%, 4.42% and 3.40%. For an IDR 100 million position, the corresponding losses are IDR 6,317,543.69, IDR 4,417,017.48 and IDR 3,403,854.00.

The study also includes quarterly earnings per share (EPS) from 2021 to 2025. Annualised EPS increases from 65.65 to 162.21. This supports the authors' argument that strong fundamentals and Sharia screening leave daily loss risk unanswered, giving investors a reason to calculate it explicitly. The study contains no strategy or P&L. It produces three thresholds.

BRIS earnings grew 2.5x; daily returns covered 23 points

Annualised EPS increased about 2.5 times from 2021 to the 2025 run-rate, following the sequence 65.65, 92.35, 123.65, 151.88, 162.21. Daily returns nevertheless covered roughly 23 points. The authors write that the EPS evidence "should be interpreted as contextual rather than causal." Elsewhere they call the two "conceptually and empirically distinct outcomes." That second claim has no test behind it.

The comparison weakens on closer inspection. Earnings begin in 2021; the price sample begins in November 2023. The 2025 estimate doubles two quarters (40.73 plus 40.37). Each quarter falls below Q4 2024's 41.16, while trailing twelve months EPS is 159.39. Earnings did increase, from 123.65 in calendar 2023 to 159.39 over the trailing twelve months at Q2 2025. The figures adequately illustrate that good earnings can coexist with large daily price moves. Calling the two "empirically distinct," however, rests on one stock without a test.

What do 1,000 Monte Carlo iterations buy?

Less simulation noise. With fixed μ and σ, repeated normal draws produce sample moments that converge to the inputs. Put those converged moments into the parametric formula and the answer converges to parametric VaR. Averaging 1,000 runs reduces the variation introduced by RAND(). It leaves the assumed tail shape intact. The introduction describes Monte Carlo results as "more accurate", though the normal assumption sets the limit on accuracy here.

For this stock, that limit matters. The -11.11% return lies about 4.2 standard deviations below zero. Under a normal model, a move that large has roughly 1.3 in 100,000 odds on any given day. Over 474 days, the expected count is about 0.006. The sample contains at least one such loss, along with a 12.32% gain at the opposite tail. The authors acknowledge that the study "assumes normality for simulation purposes but does not test" whether the returns meet that assumption. The conclusion says extreme tail behaviour is "not fully incorporated." GARCH-based VaR and CVaR are proposed for future work.

The reported VaR exceeds the closed-form 5.98% at 99%

With μ = 0.001465 and σ = 0.026318, the closed-form calculation yields:

Those differences are 0.34, 0.24 and 0.17 points. Solve for σ using the reported VaRs and the result is about 2.78%, around 6% higher than the published 2.63%.

Simulation noise cannot account for the discrepancy. Suppose each iteration draws 474 returns, as one reading of the design suggests. The per-iteration error on σ is about 0.085 points; averaging 1,000 iterations brings it down to roughly 0.003. The implied difference is about 0.15 points, some fifty times larger. The sheet must have received something else. A different input series or estimator could explain it. The paper does not let us determine which, and none of this proves a coding error. The authors still owe readers a reconciliation.

The paper defends the normal model as "a transparent benchmark of short-term downside risk." A transparent benchmark should be reproducible from the published statistics. This one currently is not. The discrepancy happens to be conservative: it pushes the 99% estimate toward the level fat tails would suggest.

Luck makes a poor control.

Before using 6.32% for BRIS position limits

The authors suggest applying these thresholds to position limits. An IDR 100 million holding would then require planning for IDR 6.32 million at 99% and IDR 4.42 million at 95%. Two checks should precede that use. The 474 returns need a distributional test, an omission the authors acknowledge. The model also needs an exceedance count. A correctly specified 99% model should record about 4.7 breaches in 474 days; a 95% model should record about 23.7. We did not find an exceedance count in the paper. And σ comes from a single two-year unconditional estimate, making it unsuitable as a live risk number on a day when volatility has doubled.

We could not perform that backtest ourselves because we had neither IDX price data nor the bank's quarterly earnings history. Substituting a US bank would test a different issuer and would not reproduce the reported thresholds.

We would use the 6.32% after seeing a worksheet that reproduces it from the stated inputs and a breach count near 4.7 in 474 days.