Every pool of capital has to become a portfolio. Whether it is a pension fund, sovereign fund, family office or model portfolio, someone has to translate objectives into actual exposures. A key step is deciding what risks to take across asset classes and markets, sizing public and private allocations, and then deciding how to manage tilts, rebalancing, and risk management of exposures as conditions change.
Institutions organise those allocation decisions in very different ways. A pension fund board may delegate strategic decisions to a total portfolio team. A sovereign investor may manage against a reference portfolio. A wealth manager may embed asset allocation choices in their model portfolios. The governance structures, constraints and implementation mechanisms are different, but the underlying questions are similar: what exposures should the investor own, how much capital should be allocated to each, what risks are worth taking, and how should those choices change as markets and circumstances evolve?
Asset allocation decisions are getting harder, and they come round more often. The anchors of the past few decades, low inflation, falling rates and bonds that hedged equities, have given way to more active fiscal policy, geopolitical and trade fragmentation, and a technology shift whose effects are still uncertain. Regimes now shift faster than annual strategic reviews, private assets make up a growing share of portfolios, and AI is multiplying the questions asked of every portfolio.
“Asset allocation decisions are getting harder, and they come round more often.”
Yet the core systems most institutions rely on for these decisions were built for ex-post analysis: to record what is held, attribute what happened and gauge risk from history. Recording the current portfolio is not the same as deciding the next one. The few tools that do exist are insufficient and piecemeal.
Asset allocation decisions need their own analytical infrastructure. We call it the “asset allocation layer”: the Capital Market Assumptions, risk engines and factors, scenarios and portfolio construction tools that give a forward-looking map and capabilities for allocating to a diverse pool of assets across public and private markets, across regions, with a record of what was assumed and decided.
The missing layer
Most of the investment technology stack has been built for other problems. The journey from capital to an invested portfolio starts with the broad objectives, liabilities, and constraints that define what the portfolio is trying to achieve. When it comes to tech and systems, there is a gap in provision for those that explicitly guide what exposures and allocations deliver on these objectives. Industry attention has tended to skip to the design of the exposures themselves: construction of indices and factors, the active mandates for beating benchmarked exposures, and the private market deal sourcing.
Further attention then gets paid to recording positions, executing trades, attributing historical risk and performance, and managing data. These needs are real, and they are well served. This operational layer is mostly built around the book of record: portfolio management systems that keep the record of positions and trades, custodial systems that value and safeguard them, attribution tools that explain what happened, risk systems that measure the exposures a fund carries, index and data providers that supply benchmarks and prices, and, increasingly, data and AI platforms that organise an institution’s information and workflows. Each records what is held or explains what has happened.
A smaller ecosystem provides tools and services that support the decision-making of institutions, but the gap remains large for asset allocation decisions. A consultant’s assumptions are refreshed once a year and lean on historical averages. A scenario module replays past episodes or applies one-off shocks to today’s holdings. An ALM model projects thousands of paths, but does not deliver the forward-looking consistency and versatility to handle the full range of asset allocation decisions. And each provider usually sits apart from the others: expected returns come from one source, risk from another and scenarios from a third, each built on different assumptions about the world.
We use the term “asset allocation layer” for what fills this gap. It is the systems and tools that turn data into coherent and forward-looking allocation decisions. The allocation layer does not replace the layers beneath it, but rather builds on them. The other layers record and explain the past, which is observable and well defined. The allocation layer informs decisions about what comes next, which cannot be observed or defined in the same way. Long-horizon assumptions have often been treated as qualitative beliefs rather than estimates that can be built and tested, and most of the industry’s effort has gone into building investment products rather than into the allocation decisions across them. That is why this layer has been missing, and why allocation decisions have been regularly left with weak foundations.
Why it matters now
Three forces are making allocation decisions more complex and more frequent, and each raises the cost of the missing allocation layer.
Changing regimes
Over the last few decades, falling rates and low inflation made a stable allocation look sufficient, and bonds hedged equities reliably. When inflation returned, stocks and bonds fell together. The forces now shaping markets, from fiscal expansion and defence and energy spending to trade fragmentation and AI investment, can each move growth, inflation and rates on their own, so the regime can shift on more than one front at once. And when real yields rise while equities stay expensive, return objectives can become reachable with less equity risk, a window an allocation reviewed once a year can easily miss. This does not mean trading the allocation constantly. It means being able to adjust it when the case changes, rather than waiting for the next scheduled review. The implication: expected returns and risk have to stay current with markets, across horizons, and scenarios have to test what happens if the current regime ends.
The total portfolio approach
Institutional investors are increasingly moving to a model where every idea competes for the same capital. This competition is only possible if each opportunity is valued on the same forward-looking assumptions, against the same cost of capital framework, and tested against the same scenarios, each at the appropriate horizon. The implication: inconsistency between sources is no longer an inefficiency but a key source of decision risk, because the spread that decides an investment is often smaller than the error between mismatched inputs.
Private assets and Alternatives
With private assets a large and growing share of institutional portfolios, every commitment is a funding decision whose effects run for years. The implication: private and public assets have to be valued in one model, and pacing has to run on the same scenarios as the rest of the fund, or the fund commits capital it may not actually want or even be able to meet when calls arrive.
The need for an asset allocation layer is not new. Institutions have always had to make forward-looking decisions across competing risks, opportunities and constraints. What has changed is the complexity and frequency of those decisions and, with it, the cost of relying on fragmented tools and processes.
What the allocation layer must deliver
Many tools cover parts of the allocation layer. A complete layer has to deliver four things:
A forward-looking map of investment opportunities and risks
Asset allocation is a decision about what to own from today. The decision inputs should begin with current market prices and analyse future returns and future risks, not the distribution that happened to occur in the past. The output is a forward-looking distribution of outcomes across assets – public and private - and across horizons, giving investors a view of both the opportunities ahead and the risks around them.
The typical approach: extrapolating historical averages and assuming valuations mean-revert, both carry a weak track record. Many published assumptions depend on which years of history were used, and the economics behind the equity numbers often do not match those behind bonds or private assets.
Consistency by construction
Expected returns, risk and scenarios need to describe the same underlying world. Too often, investors combine long-term return forecasts, historically estimated risk and short-horizon decisions that were never designed to fit together.
A stronger framework generates returns, correlations and downside outcomes from the same assumptions and over the same horizon, with common drivers such as growth, inflation, rates and valuations. That matters because expected returns are uncertain: the distribution around the forecast, and how assets behave in the environments that matter, are part of the allocation decision. Combining outputs from separate models afterwards does not resolve differences in the assumptions underneath them.
Accountability to outcomes
If expected returns, risk, and scenarios are being used to allocate capital, they should leave a record that can be tested. Forecasts should be dated and compared with subsequent returns, so investors can see whether they actually distinguished better opportunities from worse ones. That matters because many widely used assumptions have weak realised track records: our benchmarking found equity CMAs negatively correlated with subsequent returns, while our fixed-income estimates showed materially stronger relationships than major published alternatives.
Most assumption sets provide little comparable evidence at all. Producing estimates across horizons also creates a faster feedback loop, allowing shorter-term outcomes to provide evidence about the same framework before a ten-year forecast has run its course.
Traceability and explainability
Allocation decisions need to be understood well enough to approve, challenge and defend. Every output should trace back to the assumptions and economic drivers that produced it — not simply come with a plausible narrative after the fact.
Starting from what markets are pricing provides a useful reference point, separating observable market expectations from the active views an investor chooses to take. This also exposes the trade-offs that matter: if a return objective exceeds what the market appears to offer, the choice between accepting a lower return, taking more risk or changing the objective becomes explicit. Under fiduciary governance, that traceability is essential when decisions are made and even more important when outcomes disappoint.
AI and the asset allocation layer
AI will change how allocation decisions are made. An agent can already compare portfolios, run scenarios and draft a committee paper in minutes. As these tools spread, allocation decisions can shift to a much higher frequency: more people in an institution can question the portfolio, and more questions can be asked.
Each answer is only as good as what the agent reasons over, however. Speed does not improve the inputs or make the underlying calculations more reliable. An AI-enabled pipeline built on stale or inconsistent assumptions merely produces weak answers faster, at greater scale and possibly with more convincing prose. Committees will trust AI-assisted analysis only when the numbers behind it are forward-looking, consistent, tested in code and traceable to their assumptions.
Trust also needs a record, and verification is not easily compressed by AI. AI speeds up the development of systems, ours included. It does not, however, supply the judgement about what to model and the discipline of being checked. While a model trained on market history may already know how a backtest ended, a system built today needs outcomes to arrive and to have been used in the real world to earn trust.
AI applied by specialists, on an engine built and validated first, is better placed for exploiting the power of the technology. Each new capability rests on outputs that are designed to be consistent, tested and on-the-record. That is why we built the layer first and why the next step is to open it to AI agents directly. The more agents an institution runs, the more it matters what they are querying.
“The more agents an institution runs, the more it matters what they are querying.”
Built by allocators
We are building the layer as allocators, for allocators. The founding team worked together for a decade in the asset allocation team at NBIM, the manager of Norway’s sovereign wealth fund, combining macro-finance research with the mandate, governance and time to put it into practice. Building the layer takes that combination, and few institutions can assemble it in house.
We focus on R&D to advance the modelling frontier, publish forecasts that can be tested against outcomes, and have deployed our systems with reserve managers, pension funds and asset managers to demonstrate their use in live investment processes.
Questions to ask of your current tools
Whatever systems an institution uses for their asset allocation decisions, we think several questions need to be asked:
- Are expected returns aligned with what markets are pricing today?
- Do returns and risk for every holding, public and private, including external mandates the institution holds, come from one consistent model?
- Have the forecasts been published and compared with what subsequently occurred?
- Can you trace a scenario through the whole portfolio over the years that follow and show whether the return target or benchmark is still met?
- Can the relevant governance body see which driver moved a number, and find last year’s decision with the assumptions underpinning it?
If the answers are weak or ambiguous, the institution is probably trying to make decisions about the future with tools built primarily to describe the past.
