Cathie Wood is on CNBC explaining her genius investment strategy. Ron Baron is on CNBC explaining his. ARES is up 28% annually over ten years. APO is up 24%. ARKW beats the S&P by nearly nine points a year. Give them your money.

The raw numbers are real. The outperformance is not. It decomposes into three mundane things: elevated beta (every ticker here carries 1.3–1.8× the market risk of a plain index fund, which earns a risk premium in bull markets whether or not anyone is skilled), survivorship (running many high-volatility funds guarantees a spectacular right tail to advertise), and, for part of the lineup, the fee business itself, because several of these tickers are not funds at all. Strip those out and there is nothing left.

Step 1: the raw numbers

Here is what you see on CNBC. Annualised total returns, Jan 2015 through Dec 2025 (132 monthly observations; ARKF from its Feb 2019 launch), vs SPY at 13.4%.

One correction to the framing before going further: this list mixes two different kinds of ticker. The ARK ETFs and BPTRX are funds; buying them buys the manager’s portfolio. APO, BX, KKR, CG and ARES are not funds. They are the asset managers’ own publicly traded shares. Buying APO does not buy you exposure to Apollo’s funds; it buys Apollo’s fee stream, the business of managing other people’s money. Whatever those five tickers did over the decade, it measures the growth of a fee-collecting business, not anyone’s ability to pick investments. Keep the two groups separate as you read.

Fig. 1. Raw annualised returns for each ticker, sorted descending. Dashed line = SPY (13.4%). Eight of eleven beat the index on raw returns (Carlyle ties it). Several by a wide margin.

Step 2: strip the beta

Every ticker in this sample runs with a true OLS beta against SPY of 1.30–1.77. The S&P 500 carries an equity risk premium; higher beta amplifies it. That amplification is not skill; it is compensated risk. You could replicate it by borrowing and buying index funds.

ARKK’s self-reported beta is 0.89 against its own benchmark. Against SPY it is 1.77. Once you charge each ticker for its actual market exposure using Jensen’s alpha, the picture changes completely.

Fig. 2. True OLS beta against SPY for each ticker. All carry substantially more market risk than the index; the dashed line marks beta = 1.0. Every beta is significantly above 1 (t ≥ 3.3) except ARES (t = 1.5).
Fig. 3. Jensen’s alpha (annualised) after adjusting for each ticker’s true beta against SPY. The impressive raw returns of Fig. 1 shrink toward zero, and no estimate clears conventional statistical significance. The largest, ARES (+13.3%, t = 2.0), is an asset-manager stock, not a fund; see the regression table for detail.
Ticker Raw CAGR True β R² Beta explains Jensen’s α t(α)
Funds: buying the ticker buys the manager’s portfolio
ARKW Next Gen Internet22.1%1.670.53+7.7pp+4.2%0.50noise
BPTRX Baron Partners21.4%1.550.54+6.3pp+4.0%0.70noise (Tesla)
ARKQ Autonomous/Robotics18.1%1.400.59+4.6pp+1.8%0.32noise
ARKK Innovation14.0%1.770.52+8.8pp-3.1%-0.38negative, n.s.
ARKF Fintech, from 201912.8%1.740.55+8.5pp-6.4%-0.54negative, n.s.
ARKG Genomic Revolution3.6%1.620.45+7.1pp-10.5%-1.21negative, n.s.
Asset-manager stocks: buying the ticker buys the fee business, not the funds
ARES28.0%1.300.37+3.4pp+13.3%1.97borderline
APO Apollo Global23.7%1.550.46+6.3pp+7.0%1.02noise
BX Blackstone20.7%1.510.50+5.8pp+4.2%0.63noise
KKR19.5%1.670.55+7.7pp+1.7%0.29noise
CG Carlyle13.4%1.750.49+8.6pp-3.0%-0.43negative, n.s.

OLS of monthly excess returns (3-month T-bill) on SPY excess returns, Jan 2015–Dec 2025, 132 monthly observations (ARKF: 82). t-statistics use Newey-West standard errors, 3 lags. Jensen’s alpha annualised from the monthly intercept. Beta contribution = (β − 1) × realised market risk premium (11.5pp).

Eight of eleven tickers beat the S&P 500 on raw returns. After charging each for its actual market exposure, not one shows Jensen’s alpha statistically distinguishable from zero at the 5% level. The closest, ARES (t = 1.97), is not even a fund: it is an asset manager’s own stock, and its decade measures the growth of Ares’ fee-generating AUM, not anyone picking securities well.

Step 3: what’s left is survivorship

After stripping beta, the fund-side point estimates that remain positive are small and noisy: ARKW (+4.2%, t = 0.5), BPTRX (+4.0%, t = 0.7), ARKQ (+1.8%, t = 0.3). None is statistically distinguishable from zero. But even if they were real, a zero-skill manager running 8 high-volatility funds through the 2015–2025 market should expect the best fund of the family to return about 29% annually, roughly 16pp over the S&P, just from the right tail of the return distribution. You launch many, advertise the winner, quietly close the rest. ARKW’s realised 22.1% sits at the 27th percentile of that zero-skill best-fund distribution: nearly three quarters of no-skill fund families would have produced a better flagship. ARKG is the same strategy, different seed. This mechanism is old enough to have a canonical book: Taleb’s Fooled by Randomness. It is also the engine behind every backtested stock-picking service, a genre I took apart in insidercopytrading.com is a scam.

Fig. 4. Expected best-fund and median-fund CAGR from 30,000 zero-alpha Monte Carlo simulations: beta 1.5, idiosyncratic volatility scaled to the family’s total volatility, market months bootstrapped from realised 2015–2025 SPY returns. A manager with no skill whatsoever should expect their best fund to beat the S&P by 16–18pp annually just from running a large enough fund family through this bull market.

AUM makes it worse: capital chases recent winners. ARK peaked at ~$51B in early 2021, right after a 152% annual return and right before the collapse. The average retail investor bought near the top. Equal-weighted average since-inception CAGR across the eight ARK ETFs: 10.6%, below the S&P. And the dollar-weighted return investors actually experienced is lower still, because most of the capital arrived after the 2020 run-up and sat through the drawdown.

Fig. 5. ARK funds by since-inception CAGR (x-axis) and Jensen’s alpha (y-axis); bubble size = peak AUM in $B. Six of eight sit below the zero-alpha line, and the two above it (ARKW t = 0.5, ARKQ t = 0.2) are indistinguishable from it. The only ARK fund whose alpha is statistically significant is PRNT, at -13% a year (t = -2.5). The flagship (ARKK, $28B peak AUM) is negative; most capital flowed in near the top, after the return was already earned.

Appendix: private markets, same tricks better hidden

The listed shares above are the GP side of firms like Apollo and Ares. On the LP side, in the funds these firms actually run for clients, the games are the same but better hidden.

NAV smoothing Quarterly appraisals lag public markets 1–2 quarters. During COVID, PE NAVs fell ~15% while public markets fell ~35%. Measured beta: ~0.4. True beta: likely 1.5+.
Buried leverage Funds use 50–65% debt at portfolio company level. 2x leverage implies ~1.7x effective market beta, hidden in portfolio company financials.
IRR gaming Early distributions inflate IRR. Fund I was small and easy to deploy. Fund VIII is $20B in a crowded market. Advertise Fund I's 35% IRR.
GP co-invest skimming Best deals go to GP co-invest vehicles at lower fees. Fund LPs get average deal quality. Performance reported often includes GP co-invest returns.

Phalippou (Oxford, 2020): PE funds since 2006 deliver net PME of ~1.0 vs the S&P. Gross PME exceeds 1.0, but 2% management + 20% carry consumes all of it. The entire fee stream is a transfer from LPs to GPs with zero aggregate alpha. [Phalippou & Gottschalg 2009; Phalippou 2020, “An Inconvenient Fact”]

Verdict

The outperformance that gets managers on CNBC decomposes as follows. Most of it is beta: higher market exposure, higher expected return in a bull market, nothing to do with skill. What remains after beta adjustment is statistically indistinguishable from zero for all eleven tickers; the largest t-statistic in the sample is 1.97. And the two largest residual point estimates (ARES, APO) are not funds at all but asset-manager stocks, whose decade reflects the growth of fee-generating AUM and the operating leverage of the asset-management business. That is a bet on the fee machine, not evidence that anyone inside it picks investments well. The fund-side survivors with positive point estimates (ARKW, BPTRX, ARKQ) are well within what zero skill produces across a large enough fund family.

Throughout all of this, ARK collected 0.75% annually. Baron collected 1.0–1.4%. On tens of billions of AUM. On zero net alpha. That is the business model.

Strip the beta. Account for the survivors. There is nothing left except the fee.
Data: Yahoo Finance monthly adjusted close (dividends reinvested), Jan 2015–Dec 2025, 132 monthly returns; ARKF from Feb 2019; Fig. 5 uses each fund’s full since-inception history. Risk-free rate: 3-month T-bill (^IRX), monthly average. Jensen’s alpha: OLS of monthly fund excess returns on SPY excess returns; t-statistics from Newey-West (HAC) standard errors with 3 lags. Survivorship simulation: 30,000 Monte Carlo paths per family, zero true alpha, beta 1.5, idiosyncratic volatility scaled to the stated family volatility, market months bootstrapped i.i.d. from realised 2015–2025 SPY returns. PE analysis: Phalippou & Gottschalg (2009), Phalippou (2020). AUM: ETFdb.com and company disclosures. Not investment advice.