Academic Signal #10

1. PE’s outperformance boils down to leverage and valuation smoothing

TLDR

• The average PE fund looks like levered public equity with smoothed marks. Once you adjust for leverage and appraisal smoothing, returns converge to public benchmarks while risk and correlation rise.
  

• If you want the core exposure, public tools CAN deliver it cheaply and with liquidity.

The author compiles industry data and academic work to test PE’s three big sales points: 

1. higher returns, 

2. lower risk, and 

3. diversification. 

The claim is that:

• most of PE’s “alpha” is explained by leverage and rising exit multiples, 

• while quarterly, appraisal-based NAVs suppress measured volatility and correlation. 

“Outperformance” ≈ leverage + multiple expansion

The author cites (and reproduces) the deal-attribution work showing that most buyout value creation since 2010 comes from rising exit multiples and leverage, not from operating improvement.

The main backbone is McKinsey’s StepStone-based exhibits: 67% of realized buyout returns (2010–21) from multiple expansion+leverage in the 2024 review; updated to 61% for 2010–22 in the 2025 report.

He also points to Bain’s 2024 PE report, which shows multiple expansion doing much of the work and median margin expansion near 0, consistent with the idea that operating “alpha” did not drive most returns.

Appraisal smoothing hides true risk (volatility and correlation)

The paper leans on the standard “unsmoothing” literature: Getmansky-Lo-Makarov (2004) show how illiquid assets with appraisals produce serially correlated, smoothed returns; de-smoothing lifts measured vol and beta and reduces Sharpe inflation.

He brings in Brown-Ghysels-Gredil (“Nowcasting NAVs”) who estimate unsmoothed PE NAVs and find market betas around ~1.0 for buyouts and ~1.4 for VC with annual vol of 33–40%, i.e. materially higher risk and equity-like co-movement once you strip out smoothing.

More recent unsmoothing work (Couts–Goncalves–Rossi) reaches the same conclusion: observed private-asset returns are smoothed; unsmoothing raises vol and alters correlation estimates.

What recent performance shows

The paper aggregates external sources noting that PE has struggled to beat broad public benchmarks in recent years on a net basis. 

• Financial press coverage on the FT (June 11, 2025): Private market funds lagged US large-cap stocks across 3M, 1y, 3y, 5y, and 10y horizons for the first time since 2000

• That aligns with State Street Private Capital Insights, Q1 2025: State Street’s PE index posted low single-digit returns in 2024 (All PE 5.9%, Buyout 6.0%, VC 4.1%, Private Debt 8.2%) while the S&P 500 total return exceeded 20%

How to replicate buyout-like risk and returns with public tools

Academic and practitioner work show that buyout fund returns can be approximated with small/value public equities plus modest leverage, and that unsmoothed buyout risk looks like high-beta public equity. (see AQR’s paper on Demystifying Illiquid Assets: Expected Returns for Private Equity)

Use a small/value equity base + add leverage to match target risk.

Step-by-step build

1. Pick a liquid small-cap or small-value sleeve as the base equity exposure, e..g, S&P 600 ETFs like IJR or IWM for Russell 2000 (or use ETY E-mini Russell 2000 if you want cheap leverage, one RTY contract = $50 x the index).

2. Set a target annual vol. For buyouts, set a target annual vol of 30-33% as per unsmoothed estimates. Then size leverage “L” each month as:

   1. L = target_vol / realized_vol_base

   2. Use a 6-12 month rolling realized vol on your base to keep L stable. This keeps your proxy near buyout-like risk, rather than running fixed leverage that can drift.

3. Add leverage with futures: Hold your base in ETF(s) or a simple stock basket. Add RTY futures until your total notional equals L x sleeve NAV. Keep uninvested cash in T-bills as futures collateral. Contract math: contracts = target_notional / (multiplier x index). For RTY, multiplier = $50.
   

4. Optional refinement to mimic buyout tilts. Sector adjustment: overweight industrials and consumer cyclicals, underweight utilities and energy, if you want to echo typical buyout exposures noted in bottom-up work.

5. Rebalance and risk controls

   1. Rebalance monthly or quarterly to maintain L and sector/value tilts.

   2. Set drawdown guardrails. For example, if sleeve drawdown hits 25%, lower L by 25-50% until recovered. That reflects how buyout GPs de-risk in stress.

   3. Use options sparingly: RTY options can cap tail risk during periods of max leverage. They are European style, $50 multiplier, and liquid around standard expiries.

Illustrative example

Setup

• Sleeve capital: $10M

• Base: 100% Russell 2000 Value via ETF

• Futures for leveraged exposure: RTY

• RTY index level: assume 2,100 for math. One RTY notional = $50 x 2,100 = $105,000

Target

• Estimate last-12-month vol of your base at 20% (illustrative)

• Target vol 30%

• Leverage L = 30% / 20% = 1.5

Position

• Target exposure = L x NAV = 1.5 x $10M = $15M notional

• Futures needed = $15,000,000 / $105,000 ≈ 143 RTY contracts

• Initial margin is a fraction of notional; keep the rest of the sleeve in T-bills for collateral

Ongoing

• Each month, update realized vol and reset L to keep target vol.

• If realized vol rises to 25%, L falls to 30/25 = 1.2. Recompute target notional and reduce futures accordingly.

• If sleeve drops 20%, recalc target notional off the new NAV so sizing self-adjusts.

Why this should be close

• Buyout funds select small, value-leaning firms and add leverage. Public equities with those traits, plus modest leverage, have matched or beaten buyout fund returns in replication tests before fees (Erik Stafford, December 2015). Your overlay reproduces that economic mix without lockups or 2/20.

• Unsmoothing research shows true buyout vol is equity-like. Vol targeting to low-30s puts your realized risk in the right zip code. Expect more visible drawdowns than reported PE NAVs because you are marked to market daily.

Caveats

• This is a proxy. It will not capture idiosyncratic GP alpha, governance work, or deal-specific convexity. It targets the median buyout risk/return.

• Taxes: futures are Section 1256 with 60/40 treatment and year-end mark-to-market. That is a real annual drag in taxable accounts.

Bottom line

PE is not magically different. It is mostly equity risk with leverage and a long reporting lag. The paper’s advice is: measure private allocations against the right public proxies, de-smooth the inputs you hand to your optimizer, and reserve illiquidity for proven edge.

2. Funds with differentiated holdings outperform those with similar holdings by 1.2% per year

TLDR

• A simple overlap metric built from mutual funds’ holdings predicts which active funds do better. Low-overlap funds beat high-overlap funds by ~1.2% per year

The authors studied 1,758 actively managed US equity mutual funds from Dec 2005 to June 2022, using Morningstar holdings and CRSP returns. They computed each fund’s “average similarity” to all other funds using cosine similarity (explained below) of portfolio weights, then tested whether that overlap predicts future risk-adjusted returns.

The authors sorted funds into 10 portfolios each quarter by overlap and held for the next quarter. Net of fees, the low-overlap minus high-overlap spread is 0.089-0.109% per month (about 1.1-1.3% per year).

What is “cosine similarity”?

Cosine similarity is a scale-free score of portfolio overlap: the normalized dot product of two funds’ weight vectors. In long-only land it runs 0 to 1, where 1 means identical holdings and 0 means no shared names.

Let’s take an example, where Fund A is 60% AAPL and 40% MSFT, and Fund B is 70% F and 30% AAPL.

First, we define the weight vectors over the union of tickers [AAPL, MSFT, F]:

• Fund A = [0.60, 0.40, 0.00]

• Fund B = [0.30, 0.00, 0.70]

Next, we calculate the “cosine similarity” = (A·B) / (||A|| * ||B||). That is, the dot product of the 2 vectors divided by the product of the vector lengths.

1. Dot product: 0.60 x 0.30 + 0.40 x 0 + 0 x 0.70 = 0.18

2. Lengths: 

   1. ||A|| = sqrt(0.60^2 + 0.40^2) = sqrt(0.52) ≈ 0.7211

   2. ||B|| = sqrt(0.30^2 + 0.70^2) = sqrt(0.58) ≈ 0.7616

3. Cosine similarity: 0.18 / (0.7211 * 0.7616) ≈ 0.328

What does this mean about “skill”

The authors built a “Seeking Alpha” measure that asks: when a manager increases a stock, does that stock earn a positive risk-adjusted return next period (and vice versa when they cut it)? Funds with lower overlap have higher Seeking Alpha.

The higher skill of low-overlap fund managers was confirmed by the classic Daniel–Grinblatt–Titman–Wermers framework. Low-overlap funds show stronger characteristic selectivity and timing. Translation: they tilt toward the right kinds of stocks at the right times.

Manager structure also shows up in overlap

Overlap changes when managers change, which suggests it is manager-specific. On structure, teams generally overlap less than solo managers, but there is a sweet spot: teams of 2-4 have lower overlap, while teams of 5+ converge back to solo-like overlap (coordination costs).

Why this matters for allocators and PMs

Crowding is a cost. If your active dollars own the same names in similar sizes as everyone else, alpha gets competed away and fees bite harder. The overlap signal is simple to compute from public holdings and delivers clean, out-of-sample predictability across models, gross and net of fees, and with persistence.

Implementation playbook

1. Measure overlap quarterly. Use 13F or vendor holdings, compute cosine similarity versus the active-universe each quarter, and track each fund’s average similarity.
   

2. Build a low-overlap sleeve. Within your active lineup, tilt weight to funds in the lowest similarity deciles. Expect a 1.1-1.3% annual net-alpha spread over high-overlap peers if past relationships hold.
   

3. Watch team size. All else equal, 2-4 person teams show lower overlap than solo or larger teams.

Caveats

This analysis is based on US active equity from 2005-2022. The effect is measured on risk-adjusted alphas and is strongest as a spread (low vs high), not a promise that every low-overlap fund will beat the index after fees. Fees still compress average net alphas toward zero, which is exactly why avoiding crowding matters.

Key takeaways

• Overlap is a clean, holdings-based crowding gauge that forecasts returns

• Low-overlap funds add 1.1-1.3% annually of alpha relative to high-overlap peers

• The signal links to real skill and persists for up to 4 quarters

• Team design matters: 2-4 person teams tend to run less crowded books