ISSI and IHSG remain indistinguishable on tariff exposure over 60 months

An Indonesian Sharia book earns no diversification credit against US tariff risk from this paper. Oktarici and Yanthi also acknowledge that their headline exposure estimate fails to hold up across tests.

Building the volatility comparison

The proposed mechanism follows the Pástor and Veronesi risk-premium story. Tariff uncertainty broadens the range of expected cash flows and discount rates. Investors rebalance simultaneously on differing views, producing greater dispersion in daily returns within a month.

Oktarici and Yanthi examine the Indonesia Sharia Stock Index (ISSI) and the composite IHSG from January 2021 to December 2025. Each index contributes 60 monthly observations, constructed from 1,207 daily log returns. The dependent variable takes the log of the monthly standard deviation of daily returns. Their explanatory variable, zTPU, transforms the Caldara et al. trade policy uncertainty index to ln(1+TPU), then standardizes it using its 2021-2025 mean and SD. As the authors acknowledge, the index measures news coverage of trade uncertainty. It counts tariff headlines rather than tariffs.

For each index, they run OLS with lagged log volatility and one-month-lagged zTPU. M3, the main specification, also includes lagged ln VIX, the lagged S&P 500 return and the lagged rupiah change. They use Newey-West inference with three Bartlett lags. A paired moving-block bootstrap adds 2,000 draws in six-month blocks, keeping the two indices' months together when resampling.

ISSI's M3 loading on zTPU is 0.1058 (HAC p = 0.0153); IHSG's is 0.0907 (p = 0.0347). Exponentiating gives roughly 11.16% more volatility for ISSI and 9.49% for IHSG per standard deviation of transformed TPU. The respective adjusted R² figures are 0.2291 and 0.2937.

How separate are these risk pools?

The descriptive figures suggest considerable shared risk. ISSI's mean monthly volatility is 0.835% a day, against 0.839% for IHSG, and the raw volatility series have a correlation of 0.948. The authors themselves emphasize the overlap in constituents. With correlation at 0.948, the indices behave much like a single portfolio with some names screened out. Any test of differential sensitivity thus begins with two series correlated at 0.948.

The headline loading weakens under further tests

The bootstrap and differencing mostly undo the result. The abstract acknowledges this. Although the M3 coefficients pass the 5% threshold under HAC, their bootstrap intervals span -0.0983 to 0.3009 for ISSI and -0.1260 to 0.2777 for IHSG.

Persistence in the regressor is the likely explanation. The ADF test for zTPU cannot reject a unit root (p = 0.8729), while KPSS rejects stationarity (p = 0.0480). Both indices' log volatility is stationary, with ADF p = 0.0090 and 0.0233. Regressing a stationary series on a persistent one can yield a significant slope simply because the persistent series rises during a few volatile months. HAC adjusts standard errors for serial correlation in the residuals. As the authors explicitly state, it leaves the regressor's own persistence untouched.

That persistent stretch is concentrated in time, too. Raw TPU ranges from 33.893 to 1,151.356, reaching its peak in April 2025. The most influential M3 observation is May 2025.

The authors respond with S1, a specification that retains volatility in levels and takes first changes of zTPU and ln VIX. They disclose that they chose S1 after seeing the diagnostics. Its TPU loadings shrink to 0.0092 (p = 0.9385) and 0.0501 (p = 0.7212). Taking differences of every variable produces 0.0003 and 0.0320. They reject H1 because the positive link "is not consistently supported across specifications and inference methods."

Dropping 2025 supplies one counterpoint. In that sample (n = 48), IHSG's loading increases to 0.1446 and remains significant under HAC (p = 0.0425); ISSI's does not (p = 0.1244). The composite's level result therefore extends beyond the April 2025 shock alone. I would put little weight on it: the sample covers 48 months, and HAC and the bootstrap already disagreed in the full sample.

Test the gap itself

Earlier Indonesian research, including Indarta and Setiawan on 2019-2025, reported Sharia resilience to global uncertainty. Oktarici and Yanthi warn that comparing p-values across separate regressions can mislead. Instead, they bootstrap the ISSI-minus-IHSG coefficient gap directly, preserving same-month pairs. This is the right comparison. It is also the part of the paper worth knowing about.

The answer tells us little.

M3 produces a gap of 0.0151 and a 95% interval running from -0.0657 to 0.0919. The point gaps across M1 to M4 range from 0.0030 to 0.0270; all their intervals include zero. S1 gives a gap of -0.0409 within an interval of -0.1493 to 0.1483. Using three-month and twelve-month blocks makes little difference to that interval (-0.1416 to 0.1237 and -0.1374 to 0.1701).

IHSG's own loading of 0.0907 puts the M3 bounds in perspective. Holding that loading fixed, the interval allows ISSI's loading to be roughly a quarter of the composite's, or about double it. With six-month blocks drawn from 60 months, each resample contains about ten blocks. Even the endpoints are noisy.

The paper calls this outcome "reject H₂", loose language for a failure to detect a difference. The authors follow it with the correct interpretation: "this result is not an equivalence test; the absence of a detectable difference does not prove that the sensitivity of the ISSI and the JCI is identical." Their explanation deserves more weight than the hypothesis verdict.

No diversification credit for Sharia screening

The trading implication is limited. An ISSI position should receive no offset against IHSG on the assumption that screening removes exposure to trade headlines. Neither the 0.948 correlation nor the gap intervals supports that treatment. Booking the positions as identically exposed would also go beyond the evidence.

S1 gives more consistent evidence of volatility persistence. Its lagged log volatility coefficient is 0.5083 for ISSI (p = 0.0005) and 0.5228 for IHSG (p = 0.0010). The corresponding M3 coefficients are smaller: 0.2798 for ISSI (p = 0.0912) and 0.3617 for IHSG (p = 0.0205). Within S1, lagged volatility is significant (p = 0.0005 and 0.0010), whereas the TPU change is not (p = 0.9385 and 0.7212). These figures estimate exposures. We found no evaluation of a portfolio, signal or cost assumption in the paper.

We could not reproduce the analysis ourselves. We lack ISSI and IHSG price histories, and our data do not include the Caldara TPU index. Substituting news sentiment would change the regressor. We also lack the JISDOR rupiah fixing.

The authors' proposed future comparison could change my view: ISSI against a non-Sharia portfolio without shared names, using a sample long enough to cover more than one tariff regime.