Into the MATRIX: positioning CAMERA for success

The following is an extract from our Q2 2026 Asset Allocation outlook.

In previous blogs we introduced our Capital Markets Expected Returns Assumptions (CAMERA) framework for setting expected returns. CAMERA integrates return estimates from two sources: longer-run expected returns based on markets being in equilibrium and shorter-term returns based on current valuation signals.

The Strategic Asset Allocation (SAA) of a fund is based on long-term estimates of return and risk. However, we may wish to deviate from those allocations based on shorter-term signals.

In this piece we introduce our Multi Asset Tilting with Risk-aware Integration eXplorer (MATRIX) framework[1]. This was recently developed to calculate potential dynamic tilts away from the SAA. We outline five key features of our model and then provide some illustrative results.

Feature 1: Watching the green code scroll MATRIX uses current market signals.

Return signals are based on valuation metrics and how ‘cheap’ or ‘expensive’ an asset class is relative to its own history. Risk signals reflect current market volatility relative to history.

The model uses short-term (less than one year) signals for both return and risk. This is appropriate for tilting to achieve alignment with the potential trading frequency. Contrary to popular opinion, valuations matter most in the short run, at least in terms of their impact on expected returns.

Feature 2: Dodging bullets MATRIX is risk aware. 

As we explained in a previous blog, when tilting a portfolio it’s important to allow for changes in risk, not just expected returns. We use estimates for changes in risk from various sources including.

• The Cboe Volatility Index (VIX) as a guide to the volatility of equity exposure
• The ICE BofA US Bond Market Option Volatility Estimate (MOVE) index as a guide to the volatility of government bond yields
• Credit spread levels as a guide to their volatility[2]
• Recent realised return volatility

Feature 3: Not all assets bend the same way

The expected returns of different asset classes have different sensitivities to their yields. We capture this by performing regressions of historic returns against starting yields, but make adjustments to avoid a statistical bias. For example, if you look at credit, mean reversion of spread levels can mean high sensitivity of short-term returns to the level of spreads.[3] In contrast, the sensitivity of expected equity returns to the earnings and dividend yields is relatively modest.

Feature 4: There is no independence

At its core, our approach adapts a BlackLitterman model. The mechanics involve first expressing the SAA as the solution to a mean-variance problem. Return and risk signals are used to nudge the means and variances, and then the portfolio is re-optimised (subject to constraints) to form a tilted portfolio.

There are various nuances involved, including:

• The size of the ‘nudges’ i.e. how much we scale signals down before applying them, reflects a Bayesian approach
• Covariance estimates come from our in-house economic scenario generator, but we apply shrinkage to pairwise correlations. The motivation for this is that it expresses the SAA as an output of Enhanced Portfolio Optimisation (EPO) as that is a better mimic of our SAA process than standard mean variance optimisation (MVO)[4]
• We capture that less aggressive SAAs should have smaller tilts. This is achieved in the step that allow us to express the SAA as an EPO solution[5]

Feature 5: When the Oracle adds her view

Strategists can form views based on more than valuations yields or risk measures, and this may be expressed as a score. Our strategists use numbers ranging from -3 to +3. Our MATRIX framework can take these scores as inputs instead or use a blend of them with CAMERA inputs.

Mind the glitches

We’re keen to stress that strong caveats apply to the outputs of MATRIX: it’s an imperfect model that depends on many assumptions and ignores some real-life challenges.[6] As such the tilts are intended only as a suggestion. The MATRIX can guide our fund managers, but the final choice is always theirs. That said, what do they look like now?

Tilts: reloaded

Starting with an example SAA, we calculated the impact of applying MATRIX as at 28 February 2026. The charts on the previous page show the example SAA and tilts based on using both return and risk signals:

The output suggests:

• Increasing total investment and reducing cash holdings
• Overweighting government bonds and underweighting credit
• Some smaller tilts within equities

The largest tilt is away from credit and into government bonds. This reflects the low credit spread environment. It’s consistent with a broader theme we observe with MATRIX and we see larger tilts for credit than for equities. This reflects more variability in its valuation yields[7] relative to average excess returns, and greater sensitivity of expected returns to those yields.[8]

And with that, we step out of the MATRIX… at least until next quarter.

The above is an extract from our Q2 2026 Asset Allocation outlook.

 
[1] The acronym is a little contrived, but we couldn’t resist.
[2] The volatility of spreads is broadly proportional to their level.
[3] With a higher sensitivity the higher the credit spread duration.
[4] Because it produces more balanced portfolios than MVO. It also results in more stable tilts.
[5] Rather than fixing risk aversion we fix the equity risk premium and allow other expected returns and the risk aversion parameter to flex.
[6] Such as transaction costs.
[7] Spreads for credit, dividend and earnings yields for equity.
[8] Albeit this is slightly offset by a tighter coupling between risk and return for credit. The value of an investment and any income taken from it is not guaranteed and can go down as well as up, and the investor may get back less than the original amount invested.