Investment Beliefs and the Portfolio Allocations of Institutional Investors

Key Takeaways

Portfolio allocation decisions depend on investor beliefs about future returns, risks and diversification benefits, yet those beliefs, measured through their consultants’ capital market assumptions, appear only weakly linked to the actual allocations of U.S. public pension plans.
This link between beliefs and allocations becomes much stronger once real-world constraints, benchmarking incentives and varying levels of risk aversion are incorporated, with beliefs explaining between 47% and 93% of observed allocation differences.
The median pension fund would be willing to pay 0.13% to 0.30% per year, roughly one quarter to one half of average annual investment management costs, to hold portfolios more closely aligned with their beliefs, while remaining gaps likely reflect other objectives and implementation frictions.

Catch up on the webinar recording here.

Summary

Motivation: Investor beliefs about future risk and returns are central to portfolio allocation decisions. Pension funds and other institutional investors rely on long-term capital market assumptions (CMAs) from consultants, including expected return, volatility and correlation forecasts, to shape their asset allocation decisions. However, these CMAs appear only weakly linked to actual allocations, despite evidence that institutional investors form sensible return expectations that are countercyclical and positively linked to risk. This paper shows that investor beliefs, proxied through consultant CMAs, are strongly linked to pension fund allocations after appropriately accounting for the frictions that shape institutional investing.

Methodology: The authors focus on U.S. public pension plans, building a sample of 181 funds from 2001 to 2023 with data from the Public Plans Database, annual financial reports, pension fund websites and hand-collected records. Allocations are grouped into six asset classes: cash, fixed income, public equity, private equity, real estate and hedge funds. Since funds’ beliefs are not observed directly, the authors use CMAs published by their consultants before the allocation date as proxies. These CMA inputs are used to construct belief-implied mean-variance efficient allocations, which are then compared with actual allocations across asset classes, across pension funds and over time through regression and variance decomposition analyses. The comparison starts with a frictionless model, then incorporates no short-sale and leverage constraints, a tracking-error penalty against a 60/40 benchmark and varying risk aversion.

Findings:

Under an unconstrained mean-variance model, consultant beliefs do not meaningfully translate into actual asset allocations, with slope coefficients of just 0.01 across asset classes, 0.02 across pension funds and 0.01 over time, compared with a theoretical benchmark of 1 if beliefs fully pass through to portfolios.

Adding no short-sale and leverage constraints limits the extreme positions produced by the unconstrained model and strengthens the belief-allocation link, with slope coefficients rising to 0.15 across asset classes, 0.16 across pension funds and 0.03 over time.

Once a 60/40 benchmarking incentive is included, the relationship strengthens further, with slope coefficients rising to 0.26 across asset classes, 0.26 across pension funds and 0.06 over time. Accounting for the varying risk aversion of funds again increases coefficients, to 0.33, 0.30 and 0.16, respectively.

Although actual allocations move less than one-for-one with belief-implied allocations, consultant beliefs still explain much of the overall allocation pattern: variance decomposition attributes 93% of allocation differences across asset classes, 88% across pension funds and 47% over time to consultant beliefs.

Reverse causality can arise in two forms. Funds might select consultants whose published views already match their portfolios, but that is hard to reconcile with the finding that the belief-allocation link strengthens, rather than decays, as fund-consultant relationships mature. Consultants might instead adjust their assumptions toward individual clients over time, and here the evidence is suggestive rather than decisive: public pension funds account for 17.3% of the average consultant’s clients, and consultant forecasts are unbiased in aggregate.

The remaining gap between actual and belief-implied allocations is economically meaningful. At the median, pension funds would be willing to pay an estimated 0.13% to 0.30% per year to move to the belief-implied allocation, roughly one quarter to one half of average annual investment management costs.

Q&A Highlights:

Additional objectives: The allocation gap partly reflects pension fund objectives beyond mean-variance optimization, including hedging and liability management needs; accounting for these objectives through fund characteristics such as funding ratio, net cash flows, fund size and in-house equity teams helps narrow the gap.

Rebalancing frictions: The belief-allocation link is weakest over time, likely because private assets are illiquid and target allocations are reset only about every three years.

Consultant selection: Consultant switches offer a potential way to test whether changes in consultant beliefs lead to changes in allocations, but they occur in only 6.2% of fund-year observations, making it difficult to draw firm conclusions.

Future research: Potential extensions include studying governance structures, adding corporate pension fund data and exploring adjustment costs as an additional institutional friction.