Lower risk? Not in this data, and not provable either way.

The question, and why it is not obvious

Shariah-compliant funds cannot hold interest-bearing instruments, cannot hold conventional banks or insurers, cannot use leverage beyond a narrow ratio, and screen out alcohol, gambling, tobacco and adult entertainment. Two arguments follow, and they point in opposite directions.

Theory cannot settle it. That makes it an empirical question, and the UK is the natural place to ask it: London is the largest Western centre for Islamic finance, Shariah pension inflows hit records in 2025, and yet the peer-reviewed UK fund evidence amounts to essentially one paper (Reddy et al., 2017). Almost everything else in this literature comes from Malaysia, Pakistan or the Gulf.

What the literature already said, and why it disagrees with itself

Sixteen studies, tagged by what they concluded. The disagreement in this field is not noise. It sorts almost perfectly by three things: which market was studied, whether the metric was a return or a risk, and whether the window contained a crisis.

The pattern: emerging-market studies that measure risk tend to find Islamic funds lower-risk. Developed-market studies that measure return tend to find no difference or a small penalty. Index-level studies that split by regime find Islamic portfolios look best precisely when everything else looks worst. A study's conclusion is largely determined before it collects a single price, by where it looks and what it decides to call risk.

The gap this dissertation aims at: a UK sample, recent enough to include COVID and the 2022 rate shock, run through the modern five-factor framework rather than CAPM alone, and judged on risk measures rather than mean returns.

The data, and the problem sitting inside it

Method, stated so it could be repeated

Four layers, each answering a question the one before it cannot. Nothing here is exotic; the point of using the standard toolkit is that a reader can check it against the papers it comes from.

Group comparisons use Welch's two-sample t-test: unequal variances, which is the right default when one group has three members and the other has 217. Fund-level measures are equally weighted across funds; the pooled regression is not, because an unbalanced panel weights each fund by how many months it contributed.

Result 1: the univariate comparison

Every gap points the way the hypothesis predicted: Islamic funds look slightly better on Sharpe, slightly less volatile, slightly better on Treynor and alpha. Not one of them survives contact with a t-test. Beta is the starkest: 0.9148 against 0.9149, a difference in the fourth decimal place, p = 0.9923. Whatever Shariah screening does to a UK equity fund, it does not change how much of the market the fund is carrying. Move the threshold to 10% and nothing changes, because the smallest p-value in the table is 0.21.

Result 2: six factors, three specifications

The market factor is left out of the plot above on purpose: at 0.918 it is two hundred times the size of everything else and would flatten the rest into a vertical line. It gets its own exhibit, because what it shows is worth seeing on its own.

0.918135 pooled. 0.917432 Islamic-only. 0.918144 conventional-only. Three regressions, run on samples that differ by a factor of seventy-three, agreeing to three decimal places. Whatever else screening does, a UK Shariah equity fund is a UK equity fund: it carries the same market risk, and the intervals overlap almost entirely.

What does differ is style. HML is negative everywhere: a growth tilt, exactly what removing leveraged financials from an index produces. CMA is more negative in the Islamic-only fit (−0.0043 against −0.0021), and momentum, significant in the conventional sample, vanishes entirely for Islamic funds (p = 0.899), consistent with a mandate that forbids the speculative trading momentum is harvested by. Different constraints, different style exposures, indistinguishable outcomes.

Switch to t-statistics and a second lesson appears. In the conventional-only model almost every factor clears ±1.96, not because those effects are large but because 19,313 observations will find significance in almost anything. In the Islamic-only model, on 264 observations, most of the same coefficients cannot be distinguished from zero even though they have the same signs and similar magnitudes. Sample size is doing more work in this table than economics is.

Result 3: idiosyncratic risk

If screening concentrates a portfolio, the concentration should show up as risk the market cannot explain: the scatter left over once beta has done its work. This is the measure most likely to catch a diversification penalty, and the one the "narrower universe" argument predicts most directly.

–

Result 4: the only result that came close

A 5.7 percentage point difference in the worst fall an investor had to sit through, in the direction the hypothesis predicted, at p = 0.068. At a 10% threshold this is a result. At 5% it is not. It would be easy to write it up either way, and plenty of papers would. The honest treatment is to work out what the test could have seen, which is the next section, and it changes the reading completely.

Note what the chart is and is not. Both curves are averages across funds, so they show the shared shape of the period (COVID in early 2020, the rate shock through 2022) with fund-level dispersion averaged out. That dispersion is precisely what the t-test needs in order to conclude anything, which is why a chart that looks like a clear separation and a test that returns p = 0.068 are both telling the truth about different things.

What a sample of three could ever have detected

This section is not in the submitted dissertation. It is the calculation I would insist on now, and it is the part of this work I would defend hardest, because it converts a limitation everybody writes in their final chapter into a number.

A statistical test that finds nothing has two possible readings: there was nothing there, or the test could not see. Distinguishing them takes a power calculation, asking, before looking at the answer, how large a difference this design would have caught. With three funds in one group, the answer is unforgiving.

There is a further twist, and it is the reason no amount of extra data collection would have rescued this design. Precision in a two-sample comparison is governed by the smaller group. Raising the conventional count from 217 to a million moves the detectable difference by well under a percentage point. The 217 funds are not the constraint and never were. Three is.

Reading my own numbers back

Building this page meant extracting the underlying series out of the submitted document and recomputing things. One number does not survive that, and burying it would be worse than the error.

Three independent routes through the results table land on the same impossible number. UK equity funds did not lose seven per cent a month above cash for seven and a half years, over the same window the FTSE 100 rose. The level of the excess-return series carries a units artefact: a risk-free rate on one scale subtracted from returns on another.

What that does and does not damage is worth being precise about, because the instinct is to assume everything is ruined and the truth is narrower.

So the research question survives intact and the presentation of the risk-adjusted ratios does not. The right correction is to rebuild the risk-free series on the same scale as the returns and re-derive Sharpe and Treynor; the ranking between groups would not move, because a common constant cannot reorder two groups.

This is on the page because a portfolio piece that only shows the parts that came out clean is a sales document. Finding this in your own submitted work, tracing exactly how far it propagates, and publishing the propagation map is a more useful demonstration of how somebody handles data than any result would be.

What it means

Limitations, ranked by how much they matter

What I would do differently, in order: run the power calculation before collecting anything; widen the universe to UK-available rather than UK-domiciled funds, and to European domiciles if that is still too thin; cluster the standard errors; and pre-register the regime split so that "Islamic funds are more defensive in a crisis" is a hypothesis tested rather than a pattern noticed.

References

–