Unsmoothing Private Markets

The problem - Illiquid Assets Appear 'Smoothed'

Return smoothing in private markets arises from several structural and operational factors that are inherent to these asset classes. Unlike public markets, where prices adjust continuously based on real-time trades, private funds typically update valuations only on a quarterly basis or around specific capital events. This periodic valuation process introduces a time lag between actual market movements and reported returns, creating a return stream that appears artificially stable — masking the true underlying risk.

‍

In addition, private fund managers often rely on internal valuation models, comparable asset pricing, or external appraisal processes to estimate fair values. These approaches, while necessary in illiquid markets, have the effect of diffusing sharp price swings over multiple reporting periods, further reinforcing the smoothing effect. The statistical footprint of this smoothing is serial correlation, a phenomenon where current returns are highly dependent on prior returns. This autocorrelation makes private market performance appear more stable and predictable than it would be if valuations were updated daily. By testing for serial correlation, analysts can often detect whether a return series has been artificially smoothed.

‍

The consequences for portfolio construction and risk management are significant. Smoothed returns systematically understate volatility, which in turn distorts key risk-adjusted performance metrics such as the Sharpe ratio and reduces the apparent correlations with public markets. This can lead to overly optimistic assessments of private assets’ risk-return trade-off, ultimately influencing asset allocation decisions in ways that may not fully reflect economic reality.

‍

The solution – A parsimonious but simple approach for “Unsmoothing”

To address the smoothing issue, we began by selecting key private market asset classes with over 20 years of quarterly data, sourced from well-established data providers (see appendix for full details). This long-term dataset allows for a robust statistical assessment of smoothing effects across market cycles.

‍

The first step involved testing for serial correlation across various lags to quantify the degree of smoothing in each asset class. Among the asset classes analyzed, real estate exhibited the highest and most persistent serial correlation across all tested lags. This was followed by private equity, and subsequently private debt, with statistically significant autocorrelation at the 5% level.

‍

In contrast, hedge funds showed serial correlation slightly below the significance threshold, indicating a lesser degree of smoothing. For comparison purposes, we also included US  treasuries, which—as expected—displayed no meaningful serial correlation, consistent with their liquid nature. This analysis confirms that the degree of smoothing varies meaningfully across private asset classes, with the strongest smoothing effects found in real estate and private equity — two core pillars of many institutional portfolios.

‍

Statistical analysis reveals private market smoothing effects vary significantly among asset classes, with real estate showing the strongest

‍

‍

Methodology and Application

To address the smoothing bias in private market returns, we apply a quantitative approach that directly adjusts for serial correlation in the time series. This process relies on the well-established AR(1) autocorrelation model, which represents reported returns as:

Where Rt​-1 is the reported (potentially smoothed) return, and ρ captures the persistence or "memory" of returns from one period to the next. A high ρ indicates substantial smoothing, as current returns heavily depend on past returns.

By rearranging the AR(1) equation, or by fitting a more comprehensive multi-lag regression if necessary, we remove the historical valuation dependency. This adjustment allows us to estimate an "unsmoothed" return series, computed as:

‍

This de-smoothed return series reveals higher volatility, more realistic drawdowns, and stronger correlations with public markets—providing a more economically meaningful representation of risk and return.

‍

‍

The Impact of Unsmoothing on Asset Class Performance and Risk

1. ‍Volatility increases: revealing true risk‍
2. Sharpe ratio corrects downward: reflecting realistic risk-adjusted return

‍

‍

3. Diversification benefits remain: correlations with public markets still contribute to portfolio diversification

‍

In Charts 7-8, we observe that reported correlations, prior to unsmoothing, exhibit desirable attributes for portfolio optimization. However, these correlations are subject to the effects of return smoothing. On the right-hand side, the delta after unsmoothing is displayed, showing the impact of removing serial correlation from the time series.

‍

While correlations increase in some cases, the overall low correlation persists, maintaining a positive contribution to portfolio diversification. However, this comes at the expense of a more volatile return profile, reflecting the true underlying risk of private market assets.

‍

‍

Practical Impact of Smoothing on Portfolio Basis

Our practical analysis demonstrates that portfolio allocations naturally shift toward assets with lower serial correlation when optimization input data is properly adjusted. Contrary to expectations, alternative assets maintain strong diversification capabilities and in capturing the illiquidity risk premium, thus reducing the probability of shortfall in meeting objectives due to their higher returns compared to traditional asset classes.

‍

In Chart 9, assets with higher smoothing effects (like Private Equity) receive lower allocations after correction due to their heightened volatility and reduced Sharpe ratio contributions. However, their allocation remains significant, with reallocation primarily occurring within the alternative asset category itself. On the liquid side, allocations remain stable and therefore not displayed.

‍

In other words, when using a mean-variance approach with accurate volatility estimates, assets with high serial correlation receive appropriately lower allocations unless constrained by investment policy limits. For instance, Private Debt, with its lower smoothing effects, maintains a stronger position in the optimized portfolio compared to Private Equity, which loses allocation after unsmoothing. Meanwhile, Real Estate remains unselected due to its low return contribution per unit of risk.

‍

Therefore, the most prominent effect occurs in the shape of the efficient frontier, which shifts rightward to display more realistic levels of return per unit of risk. Chart 10 provides a comparison of both allocation approaches on a full portfolio basis.

‍

To validate these findings, we conducted two separate optimization scenarios with smoothed and unsmoothed data sets. These scenarios incorporated the client's standard investment constraints and objectives while following a Non-LDI total return approach typical for European pension funds.

‍

While this approach effectively addresses certain challenges in alternative investments, investors should consider several additional factors. These include return distributions, optimization model variations, adjustments for non-constant volatility, and the various sub-strategies we incorporate in our most sophisticated models. For investors interested in exploring these advanced considerations, we welcome you to contact our team for more detailed information.

‍

Addressing smoothing effects is crucial for proper capital allocation, though the fundamental nature and benefits of private markets remain strong for institutional portfolios. Private market assets continue to reduce shortfall probability while maintaining beneficial correlation benefits at the portfolio level.

‍

‍

‍

For more information related to the foundational paper upon which this paper was written, please reach out to us at  info@klarphos.com.

‍

‍

Endnotes

Data Source: Preqin, Albourne, Klarphos, quarterly data from 2002Q1-2024Q3. The Infrastructure asset class was excluded due to insufficient historical data.

1. In this approach, we calculate each asset’s sample autocorrelation at lag 1 and assess its significance using a Student’s t-test (5% level). If the lag-1 autocorrelation is meaningful, we treat the reported series as having been smoothed by a one-period filter. We then invert that filter by removing the prior-quarter influence and renormalizing the series so its volatility matches the original level. Finally, all performance metrics (annualized return, volatility, Sharpe ratio) are recomputed on the adjusted, “unsmoothed” returns. This method is algorithmically straightforward and well suited to limited quarterly datasets where higher-order autocorrelations are negligible.
2. Portfolio optimization was conducted using a mean-variance optimization and a set of constraints typical for non-LDI allocation investment plans of European pension funds. The exercise was run twice with two different sets of data: one with default data points and one with adjusted time series incorporating the autocorrelation adjustment.
3. Investing in Private Markets. The Journal of Portfolio Management 50, no. 7 (June 2024).
4. CFA Institute. CFA Program Curriculum 2024 Level III: Core Volume 2. CFA Institute, 2024.

‍