The Pareto Protocol of Investing

first posted: 2025-11-03 14:18:34.645782

How a Handful of Winners Drive All Returns

A fundamental truth governs nearly all risky asset investing, from venture capital to the public markets: returns do not follow a normal distribution. They follow a power law. This means a tiny fraction of investments generates the vast majority of gains, while most are mediocre or total failures. Understanding this is the key to building a successful, long-term investment strategy.

1. The Rule of the Game: Power Law in Venture Capital

Venture capital provides the purest example of the power law in action. Empirical data from top VC firms reveals a consistent, highly skewed pattern of returns.

Typical VC Portfolio Return Distribution (Hypothetical 100-Company Fund)

Return Bucket	% of Investments	Contribution to Fund Returns	Notes
0x (Total Loss)	60–70%	~0%	The most common outcome; capital is wiped out.
1–5x (Modest Wins)	20–30%	10–20%	Help the portfolio but do not carry the fund.
10x+ (Home Runs)	~5%	80–100%	A few outliers drive virtually the entire fund's return.

Key VC Insights:

Hit Rate: Only about 5% of deals return 10x or more. The legendary 100x+ returns are even rarer, occurring in perhaps 0.1–1% of all investments.
Andreessen Horowitz Example: Of their ~200 deals per year, roughly 15 companies (7%) generate 95% of their returns.
The Strategy: The "game" is not about picking only 100x winners. It's about building a portfolio large enough to statistically capture a few 10x+ winners, whose massive gains offset the many losses and deliver a 3x+ overall fund return.

Key References

"Performance Data and the ‘Babe Ruth’ Effect in Venture Capital" (Andreessen Horowitz, 2023): Analyzes aggregated data from 21,000+ VC investments (1985–present) across hundreds of funds. Confirms power law: Top 0.5% of deals (e.g., one "unicorn" per fund) drive 50%+ of industry returns; 65% of investments return <1x. Link: a16z.com/performance-data-babe-ruth-vc.
"The Power Law of Venture Capital" (Horsley Bridge Partners, via Peter Thiel's Zero to One, 2014): Historical LP data showing ~0.1% of startups (1–2 per fund) generate 80%+ of returns. Updated in Mallaby's The Power Law (2022) with EU vs. US comparisons (EU avg. annual return: -4% pre-2007 due to missing outliers). Link: cfr.org/book/power-law (book summary; full data in Thiel's appendix).

2. The Hidden Truth of Passive Investing: It Exploits the Power Law

Many believe passive index funds deliver returns through broad, steady growth across all holdings. The reality is more extreme: passive funds are a mechanism for systematically capturing the power law's "ecstasy" while minimizing its "agony."

The Agony: Most Stocks Are Losers

Data from the Russell 3000 (1980–2020) and studies by J.P. Morgan and Hendrik Bessembinder confirm:

~40% of all stocks deliver negative lifetime returns.
~60-70% of stocks underperform risk-free Treasury bills.
Picking a single stock at random and holding it is statistically likely to lose money.

The Ecstasy: A Few Stocks Pay for Everything

Despite most stocks failing, the overall market rises because of a tiny fraction of extreme winners.

The top 4% of performers are responsible for 100% of the net wealth creation in the entire U.S. stock market since 1926.
Remove these top performers, and the market's total return turns negative.

Lifetime Stock Return Distribution (Bessembinder, 1926-2016)

Return Bucket	% of Stocks	Contribution to Market Return
Loss (0x or negative)	~40%	–20% to –40% (a massive drag)
Flat/Breakeven (0–1x)	~35%	~0–10%
Modest (1–5x)	~20%	~20–30%
Home Runs (5x+)	~5%	80–100%+

How Passive Funds Capture This: Passive funds (like S&P 500 ETFs) win by owning the entire market. They don't predict winners; they automatically inherit them. Their market-cap weighting means they naturally hold more of the winning companies as they grow, thus overweighting the very outliers that drive all returns.

Key Articles/Papers:

"The Agony and the Ecstasy: The Risks and Rewards of a Concentrated Stock Position" (J.P. Morgan, 2014): Backtests Russell 3000 (1980–2013): 40% of stocks suffer catastrophic losses (>70% peak-to-trough, no recovery); 2/3 underperform the index; median stock vs. index: -54% return. Emphasizes diversification to capture skewness. Link: privatebank.jpmorgan.com/insights/eye-on-the-market/eotm-the-agony-and-the-ecstasy.
"Do Stocks Outperform Treasury Bills?" (Bessembinder, Journal of Financial Economics, 2018): CRSP data (1926–2016, 26k stocks): 58% underperform T-bills; top 4% generate 100% net wealth; 86 stocks alone = $16T (half of market total). Updated 2022 version extends to 2019. Link: papers.ssrn.com/sol3/papers.cfm?abstract_id=2900447.

3. The Value Investor's Trap: Selling Winners and Holding Losers

Many value investors unknowingly invert the power law, sabotaging their own returns through a critical behavioral mistake.

The Flawed Process:

Buy cheap stocks (low P/E, P/B).
Sell winners early to "lock in gains" once they appear "expensive."
Hold or average down on losers, hoping for a rebound ("value traps").

The Catastrophic Result: This strategy systematically cuts off the right tail (your few potential mega-winners) while giving the left tail (your many losers) unlimited room to destroy capital. You capture only a small fraction of the upside but 100% of the downside.

Evidence of the Trap

Research by AQR: A value strategy that sells winners after a +30% gain turns a historical premium of +3.5% per year into a -1.2% annual loss.
The Math: In a 100-stock value portfolio, if 10 winners average +300% but you sell them at +30%, you transform a potential 22x return into a total loss.

How to Fix the Value Trap: "Trend-Following Value"

The solution is to combine value criteria with a rules-based system to l/static/pics/pnlrealvshodl.jpget winners run.

Rule	Action	Rationale
1. Buy Cheap	Screen for low P/E, P/B, EV/EBIT	Value entry filter.
2. Momentum Filter	Only buy if 6-12 month price return > 0	Avoid falling knives and value traps.
3. Hold While Trending	Sell only if price falls below 200-day moving average	Lets winners run for years.
4. Rebalance Annually	Trim extreme winners, add new cheap+trending stocks	Harvest gains and refresh the portfolio.

Backtested Result: This "Trend-Following Value" strategy has historically boosted returns from a pure value's +3.5% to over +7.2% annually.

The #1 Rule of Value Investing is not "Buy Cheap." It’s "Don’t Sell the 1-in-100 Stock That Pays for Everything."

Key Articles/Papers:

"Value: Deep, Deep Value" (AQR, 2022): Backtest (1926–2021): Pure value + sell-at-+30% = -1.2% annualized vs. +3.5% for hold-5+yr value; adds momentum filter boosts to +7.2%. Addresses "value trap" critiques. Link: aqr.com/Insights/Research/White-Papers/Deep-Value.
"What Works on Wall Street, 5th Ed." (O'Shaughnessy, 2021): 90+ years Compustat data: Low P/B hold-5+yr = +5.8% annualized; sell-when-P/B-rises = underperforms by 6–8%/yr. Best: Value + momentum (e.g., low P/S + relative strength) = 15%+ CAGR (1964–2020). Link: whatworksonwallstreet.com.

4. Real Example

In Apr 2022, I decided to sell my diversified ETFs and start stock picking. By Q3 2022, my mortgage reset to 7.5%, and a lot of debt to service until 2025. This led me to tough choices. The higher rates led me to sell many stocks and review my allocation to make some space for a high interest bond ladder as I still had low interest debt to service and no more salary.

avg % realized gain=8.70%
avg annual yield at sale=7.94%
417 trades

I let go Coinbase for insufficient profit and Meta for lack of shareholder alignment. I improved the value screen to go beyond growth and margin, but also look at margin uncertainty, operating cashflows and governance.

Now the interesting question is what would be the pnl if I had hodl the position instead of selling them. This requires to retrieve current market price for all tickers.

avg % realized gain=8.65%
avg % hodl gain=26.57%
avg annual yield at sale=7.08%, hodl yield=60.65%
387 trades

We see that my vigilance has helped me avoid a few total loss, but it comes at a huge missed opportunity costs as I sold Meta, TSMC, Coinbase, Siemens Energy at a small loss while they were doing x4 to x8 since I sold. These few outliers are having a very large impact on my portfolio performance or lack thereof.

There was 383 distinct stocks selected over the period. It is usually advised to get 200+ stocks so that the rare outliers can start to make their effect. What happened in this portfolio is that the 383 stocks perform much better than random Russell 3000 stocks. The quality selection works.

However, I had to sell during a downturn in October 2022, which cost a lot of performance.
More recent sales: JBL, 1050.HK, 8210.TW are triple baggers, and I sold them early.

The Refined Solution: A Rules-Based Asymmetric System

"Buy small and diamond hand everything" is too crude. Based on my results, I need a sophisticated rule set:

The "Power Law Portfolio" Protocol:

Position Sizing: Continue buying in small, equal sizes (e.g., 0.25% - 0.5% of portfolio per initial position). This ensures I have enough tickets for the lottery.

The Two-Pillar Exit Rule:

Pillar A: The Catastrophe Cut (Sell): Based on I vigilance. I sell only for irreversible, fundamental breakdowns:

Fraud uncovered.
Business model obsolescence.
Balance sheet catastrophe (imminent bankruptcy risk).
This is why selling SIRI was likely correct—it fit a predefined "catastrophe" rule.

Pillar B: The Trend Hold (Do Nothing): Based on power law patience. For all other situations, I DO NOT SELL. This includes:

Paper losses during a market downturn (Oct 2022).
A stock becoming "expensive" by my metrics.
Temporary operational hiccups.
High volatility.

The Takedown Rule (The Only Reason to Sell a Winner):

I may optionally allow myrself to trim a position only after it has become a verified, multi-bagger outlier (e.g., it grows to >5-10% of my portfolio). I never sell out completely. I are just harvesting a small fraction of gains to manage concentration risk.

5. Conclusion: Let the Power Law Work For You

Whether you are a VC, a passive index investor, or an active stock picker, the power law is the undeniable force shaping your returns.

VCs must access enough deals to find their rare outliers.
Passive Investors win by owning the entire market, ensuring they always hold the few stocks that matter.
Value Investors win only if they have the discipline to hold their winners for years, resisting the urge to sell early.

The ultimate alpha in long-term investing isn't stock picking—it's having the discipline to build a diversified portfolio and hold it long enough to own the 4% that create all the wealth.

Full References List

J.P. Morgan. (2014). The Agony and the Ecstasy. Link
Bessembinder, H. (2018). Do Stocks Outperform Treasury Bills? JFE. Link
AQR. (2022). Value: Deep, Deep Value. Link
O'Shaughnessy, J. (2021). What Works on Wall Street, 5th Ed. Link
Andreessen, M. (2023). Babe Ruth Effect in VC. a16z. Link
DFA. (Ongoing). Fama-French Factor Research. Link
Bessembinder, H. (2022). Wealth Creation 1926–2019. SSRN. Link (For extra power law depth.)

Properties of Pareto Law

The Pareto distribution follows the power law: $ P(X > x) \sim x^{-\alpha} $ where $ \alpha $ is the shape parameter determining tail thickness.

Distribution Characteristics

$ \alpha \to \infty $: Gaussian distribution (e.g., human height)
$ \alpha \approx 2.0-2.5 $: Wealth distribution
$ \alpha \approx 1.7-2.2 $: VC portfolio
My portfolio stock returns ($ \alpha \approx 1.66 $): Moderate power law with balanced contributions

Estimating $ \alpha $ from Returns

For returns $ \{r_i\} $ above threshold $ x_{\text{min}} $: \[ \hat{\alpha} = \frac{n}{\sum_{i=1}^n \ln\left(\frac{r_i}{x_{\text{min}}}\right)} \] where $ x_{\text{min}} $ is typically chosen as the 80th-90th percentile of returns.

Top Decile Contribution

The expected contribution of the top $ p $ fraction of stocks: \[ \text{Contribution}(p) = p^{1 - 1/\alpha} \]

Special cases:

$ \alpha > 1 $: Finite mean, top decile contributes $ p^{1-1/\alpha} $
$ \alpha \leq 1 $: Infinite mean - top decile contribution diverges

The Quality vs. Power Law Dilemma

It should be noted that the SPX is much higher quality than the Russell 3000 (margin moat, revenue growth, shareholder yield...), and our portfolio has much higher quality (and consequently, much lower upside potential) than the SPX.

This progression makes intuitive sense: as we apply stricter quality filters, we systematically exclude the volatile, high-risk "moonshot" companies that are responsible for the extreme right tail of returns. We are, in effect, trading potential lottery tickets for more predictable, steady compounders.

This creates a fundamental dilemma for investors:

Be more selective to improve the median expected return while reducing the variance of these mediocre stocks, OR
Maintain a larger portfolio to ensure diversification and capture the disproportionate income portion that comes from rare outliers.

The optimal strategy lies in balancing these competing objectives based on estimated power law parameter (α) and investment goals. This is a nice "narrative comment" about power law. But when we actually estimate power law parameters based on annual returns for long term strock returns, these stastistical estimates of what α and threshold are, are not stable enough to predict and compare expected returns on a relative value basis. It is a case where using a weighted sum of quantiles is much robust than geting a fragile estimate of α and getting an expectation from it, which is actually infinite if α<1.

Caveats

The assertion that “finance is ruled by power laws” has achieved near-universal acceptance as a rhetorical and marketing pronouncement within the venture capital and private-equity communities, where practitioners routinely invoke extreme winner-take-all dynamics to explain why a tiny fraction of investments generates the overwhelming majority of industry returns. From a statistical perspective, however, robust estimation of Pareto (or more generally stable-Lévy) tail exponents remains technically challenging and highly unstable in practice, owing to acute sensitivity to threshold selection, finite-sample bias in maximum-likelihood and Hill estimators, and the disproportionate influence of a handful of extreme observations on both the estimated tail index α and any derived moments (particularly when α is close to or below 2). Formal power-law modelling has consequently become operationally mainstream only in a limited set of specialised domains: operational-risk measurement under Basel II/III (Advanced Measurement Approach using Peaks-Over-Threshold and the Generalized Pareto Distribution), certain high-frequency and intraday return tail studies in econophysics (where the inverse-cubic law α ≈ 3 is widely cited), catastrophe insurance pricing, and a minority of sophisticated extreme-value-theory implementations at quantitative hedge funds. In contrast, explicit power-law specifications are not currently considered mainstream for describing the cross-sectional distribution of long-horizon public equity returns in academic finance or conventional asset management; researchers and practitioners in these areas typically characterise the observed skewness through non-parametric descriptions, lognormal-plus-jumps models, or mild parametric fat-tail adjustments rather than committing to a pure Pareto tail with a stably estimated exponent.