2007-08: four of five Islamic indices carried higher volatility

Rotating into Sharia-compliant index trackers before a crash gets less backing from Ahmad, Noor, Ijaz and Bhutta than their abstract suggests. They quote ARCH coefficients for just one index pair. Most of the evidence in their tables runs against the trade.

What protection should the screens provide?

The authors describe Islamic finance through its "principles prohibiting excessive risk-taking and interest-based transactions". They connect those principles to 2008 through Islamic indices' "limited exposure to speculative activities, which were highly implicated in the crisis." Their argument predicts a smaller immediate volatility response when a crisis arrives.

The study uses daily closing prices from Datastream for five conventional/Islamic pairs. These are Dow Jones Composite against Dow Jones Islamic Market, S&P 500 against S&P Shari'ah, FTSE against FTSE Islamic, Malaysia's KLCI against the FTSE Bursa Malaysia Hijrah Shari'ah, and the Jakarta Composite against the Jakarta Islamic Index. The authors fit a univariate GARCH model to each return series, separately over a COVID-19 window and a 2007-08 window. They interpret the ARCH coefficient, the weight on yesterday's squared shock, as telling "about the sudden response to volatility". The GARCH coefficient weights yesterday's variance and is interpreted as persistence. They also fit ARMA models to returns to assess "predictability".

The Dow Jones pair during COVID supplies the headline result. For the Islamic index, ARCH is 0.718 and GARCH is 0.2084 (p=0.03). For the conventional index, ARCH is 1.007 and GARCH is 0.189. The authors infer that "initial resilience of Sharia indices against abrupt circumstances is more powerful" than for conventional indices. They acknowledge that "the persistence of volatility across both Islamic and conventional indices suggests that additional risk management is essential, even with Islamic finance principles in place."

The abstract makes broader claims: "greater stability compared to conventional indices" and a "stabilizing role of Islamic finance principles". The narrower proposition deserves the test: Islamic indices absorb the initial hit better. An allocator following that proposition would hold the Islamic twin into a crisis to cushion the first leg.

Most pairs were more volatile

Dow Jones is the exception. Over the 165-day COVID window, its Islamic index recorded a standard deviation of 1.999, versus 2.712 for the conventional composite. Elsewhere, the gap disappears or changes direction. FTSE Islamic recorded 2.598 against the FTSE's 2.046. Jakarta Islamic came in at 2.242 against 1.942; the Malaysian Shari'ah index at 1.314 against 1.282. The S&P pair effectively ties at 2.544 against 2.543.

The pattern recurs in 2007-08. Standard deviations were 2.703 against 1.646 for FTSE Islamic, 2.018 against 1.851 for S&P Shari'ah, 1.339 against 1.073 for Malaysia, and 2.076 against 1.903 for Jakarta. Again, only Dow Jones has a calmer Islamic side: 1.477 against 1.729.

The authors acknowledge these results. Their discussion says FTSE Islamic "faces higher volatility with the S.D 2.703" in 2007-08, the highest standard deviation of any series in that window. They interpret the COVID figures as evidence that the pandemic "equally affected the performance of both markets in terms of high volatility." Yet they retain the conclusion that Islamic indices are more resilient. An allocator pays or benefits from the unconditional volatility experienced during the crisis window. By that measure, the Islamic label selected the riskier index in four of five pairs in 2007-08 and three of five during COVID, with the S&P pair tied.

Persistence needs alpha plus beta

The paper directs readers to Table 4 for the ARCH/GARCH estimates. We could not find it in the published text. The prose supplies only a few coefficients, one p-value and no standard errors. We also found no specification of the GARCH order or error distribution.

Even those few coefficients raise problems. The conventional Dow's ARCH coefficient alone reaches 1.007 during COVID; alpha plus beta comes to 1.196. For 2007-08, the same index totals 1.149 (0.206 + 0.943). The Islamic Dow totals 1.009 (0.249 + 0.76) in 2007-08. All imply variance without a finite unconditional level. The paper discusses none of them.

Its description of COVID volatility as "less persistent" than 2008 rests on the GARCH coefficient alone. For the conventional Dow, that coefficient is 0.189 during COVID versus 0.943 in 2007-08. Persistence in this model is measured by alpha plus beta. Using that sum, the conventional Dow has greater persistence during COVID: 1.196 against 1.149.

The claimed direction of the Islamic/conventional difference also changes. In discussing 2007-08, the paper says Islamic indices "such as RDJIMI and RFTSEI, exhibited lower ARCH values compared to conventional indices". These are the Dow Jones Islamic and FTSE Islamic indices. But the quoted Dow Jones estimates put the Islamic coefficient higher, at 0.249 versus 0.206 for the conventional index. For COVID, the paper says Islamic ARCH terms were "generally lower than the GARCH coefficients". Its Islamic Dow estimates are 0.718 against 0.2084.

Across the two crises, then, the sole pair with quoted ARCH estimates agrees with the abstract once and contradicts it once. The paper reports no formal test of the difference.

Fitting 949 days and 2008 into 2020-21

Daily closes "from January 2020 to December 2021" are the stated data on three occasions. The counts cannot fit that description. There are 165 observations in the COVID window and 784 in the 2007-08 window. Their combined 949 exceeds the trading days available in two calendar years. And a 2020-21 period cannot include 2008. We found neither a start date nor an end date for either window.

Some histories are shorter still. FTSE Islamic has 394 observations in the 2007-08 window, while S&P Shari'ah has 459. Both also show peculiar correlations with their parent indices: 0.024 for S&P 500 against S&P Shari'ah and 0.097 for FTSE against FTSE Islamic. By comparison, Dow Jones against Dow Jones Islamic gives 0.844, and the Malaysian pair gives 0.815.

The authors suggest that "unique market characteristics or investor bases" may explain the S&P Shari'ah index's low correlations. Screened subsets of the same large caps should resemble the Dow Jones and Malaysian pairs, with correlations of 0.844 and 0.815, rather than approach zero. I would check the date alignment before reading this as diversification.

Predictability stays in-sample

The ARMA results come from in-sample fits, with lag orders varying across series from AR(1) to AR(4). In two fits, AR and MA terms almost cancel, suggesting redundant extra parameters. During COVID, FTSE combines AR(1) -0.864 with MA(1) 0.975. During 2007-08, Jakarta combines AR(2) 0.878 with MA(2) -0.847. The Islamic Dow's COVID MA(2) estimate is printed as 0.268, alongside a standard error of 0.249 and a p-value of 0.000. Those figures cannot all be correct. We did not find an out-of-sample forecast evaluation or a trading rule.

We did not replicate this.

Our data lacks the pre-2010 daily history needed for the 2007-08 comparison. We also could not confirm coverage of the specific Islamic index series. Substituting plain US benchmarks would erase the Islamic/conventional distinction on which the hypothesis depends.

I would change my view if we could see the Table 4 cited in the text but missing from what we found. It would need standard errors for all ten series, the GARCH order, dated windows, and a pair-by-pair test of whether Islamic alpha is lower. Until then, the Dow Jones Islamic index's COVID standard deviation of 1.999 remains one data point. FTSE Islamic's 2.703 in 2008 remains another, pointing in the opposite direction.