Skip to main content

Ignoring base rates in forecasts

An analyst is forecasting revenue for a software company that is expanding into a new vertical. The company has never operated in this vertical before, but the company's CEO is confident about the opportunity. The analyst listens to the CEO's reasoning, examines the addressable market size, and builds a five-year forecast incorporating steady market-share gains in the new vertical.

What the analyst is not doing is asking: Of the 100 software companies that attempted to expand into a new vertical in the past 20 years, how many achieved meaningful share and profitability in that new market? What was the distribution of outcomes?

This is the base rate—the historical baseline probability of a particular outcome across a reference class. Ignoring the base rate and instead building a forecast anchored to the specific case (the CEO's story, the addressable market) is one of the most common mistakes in analyst forecasting.

Quick definition​

Ignoring base rates in forecasting occurs when an analyst builds a bottom-up model of a specific outcome (e.g., a company's penetration of a new market, or the success of a new product) without anchoring to the historical baseline probability of that outcome type across comparable cases. The analyst focuses on the particulars of the case (compelling story, large market) while neglecting the statistical track record of similar attempts.

Key takeaways​

  • Base rates are powerful: in most domains, the base rate (historical success rate of similar ventures) is more predictive of specific outcomes than detailed bottom-up models, yet analysts ignore them because they are specific, exciting, and based on fresh information.
  • Software companies that entered new verticals have a historical success rate of roughly 20–30% for generating more than 10% of revenue from the new vertical within five years. Analysts building five-year models forecasting 15–30% revenue contribution from the new vertical are implicitly assuming this company will outperform the base rate significantly.
  • The reference class matters: base rates for "software company entering adjacent vertical" are different from base rates for "hardware manufacturer entering software." The analyst must be precise about the reference class or risk comparing apples to oranges.
  • Analyst confidence about a specific case (the story, the management team, the differentiation) is almost uncorrelated with the base rate. A "best-in-class" management team still faces base-rate odds when entering a new market. Confidence and probability are not the same.
  • Tournaments or competitions where contestants have inside knowledge of the details (like the CEO has inside knowledge of the market) show that base-rate thinking beats confidence in the specific case more often than specific-case confidence beats base-rate thinking.

What base rates tell us​

A base rate is simply a historical frequency. Here are some illustrative examples from equity research contexts:

New product launches in mature industries: Of 100 major new products launched by companies in mature industries (consumer goods, industrial, etc.), roughly 40–50% reach 5% of company revenue within five years. The rest fail to gain meaningful share, are discontinued, or take longer. Yet analysts forecasting a new product launch often assume 10–20% contribution within five years, implying they believe the specific case will beat the base rate by 2–4x.

CEO track record transfers: When a CEO with a strong track record at one company joins a new company in a different industry, the new company often underperforms the CEO's previous company. The base rate for CEO track record transferring across industries is roughly 40–50%. Analysts often assume the "star CEO" will perform as well at the new company, ignoring the base rate that says track record in one context does not transfer.

Corporate venture success: Large corporations establish venture-capital or venture-builder arms to invest in or build new businesses. The base rate for corporate venture units to deploy capital and generate returns above cost of capital is roughly 15–25%. Yet analysts and executives often model corporate venture as though success rates should be 50%+, implying they are assuming far above-base-rate performance.

International expansion in new markets: A consumer company that is successful in its home market expands to a new country. The base rate for achieving comparable profitability (operating margins within 500 basis points) in the new country within five years is roughly 30–40%. Analysts frequently model 60–70% near-parity margins in new markets, ignoring the base rate.

Turnarounds and restructurings: Of 100 companies that undergo significant restructuring after deteriorating business performance, roughly 30–40% recover to prior profitability levels within five years. Another 30–40% continue to deteriorate. Yet analysts covering turnarounds often implicitly assume 70–80% success rates in their models, far above the base rate.

Why analysts ignore base rates​

The psychology of why analysts ignore base rates is well-understood:

Availability bias: The analyst is focused on fresh information (the CEO's growth plan, the market opportunity size, the differentiated strategy). This information is vivid and easy to retrieve, so the analyst weights it heavily. The base rate (the historical track record of similar ventures) is abstract and requires deliberate research to retrieve, so it is weighted lightly.

Specific case fallacy: Analysts genuinely believe that the specific case is special. "Yes, 70% of new product launches fail, but this company has the best in-class CEO, differentiated technology, and unique market position. This case is in the top quartile." The analyst is usually not wrong that the specific case has some strengths, but the analyst overestimates the degree to which those strengths insulate the company from base-rate odds. Most cases that beat the base rate also have strengths; having strengths is not sufficient to know you will beat the base rate.

Incentives to be optimistic: Analysts who forecast base-rate probabilities (70% chance of failure) are implicitly assigning pessimistic outcomes. This is career-dangerous; the analyst looks wrong if she predicts failure and the company succeeds (which happens 30% of the time). The analyst looks right if she predicts success and the company delivers, or if she predicts success and the company fails (the analyst can say "I got the narrative right, just timing was off"). The incentive structure pushes toward optimistic, specific-case thinking rather than base-rate thinking.

Narratives are compelling: Humans are wired to respond to stories. A CEO saying "We have identified a $10 billion market, we have unique technology, our team has deep domain expertise" is compelling. A base rate saying "Of the last 20 companies that attempted this, 14 failed" is not. The analyst builds the narrative and ignores the numbers.

A framework for incorporating base rates​

This framework shows how to integrate base-rate thinking with bottom-up modeling.

Real-world examples​

Google's diversification attempts (2000s–2010s) Google's base business (search advertising) was growing 20%+ annually, generating enormous cash. The company attempted diversification into multiple new verticals: smartphones (Android), autonomous vehicles, commercial robotics, life sciences (Verily). The base rate for technology companies successfully diversifying into non-core verticals is roughly 20–30%. Google's attempts have had mixed success: Android became massive; autonomous vehicles remain unproven; robotics and life sciences have not generated material revenue. Analysts who modeled each new Google venture often assumed 15–20% contribution to company value within 5–10 years, beating the base rate by 2–3x. The portfolio approach (make many bets, some will pay off) may work, but individual ventures face base-rate-level odds.

Microsoft's enterprise cloud bet (2008–2015) In the late 2000s, Microsoft began building Azure, its cloud platform, to compete with Amazon Web Services. The base rate for large, established software companies successfully pivoting to cloud infrastructure is low (maybe 20–30% of attempts become material revenue drivers within 5–10 years). Microsoft's success was not predetermined; it reflected superior execution, deep enterprise relationships, and timing. Yet analysts modeling Microsoft's cloud opportunity in 2010–2012 often forecasted Azure becoming 10–15% of revenue by 2015, which would have been above-base-rate success. When Azure grew more slowly than forecasted, it was partly because base rates matter more than specific-case optimism.

Amazon's high-margin businesses (AWS, advertising, marketplace fees) Amazon's highest-margin businesses (AWS, advertising, marketplace fees) grew from internal initiatives. The base rate for large retailers successfully launching high-margin service or software businesses is low. Yet analysts modeling Amazon often assumed these businesses would reach 20–30% of company operating income within five years, beating the base rate. That Amazon accomplished this is a genuine success story; it is not evidence that the specific-case analysis was better than base-rate thinking, only that Amazon is a genuine outlier. The base rate should have been incorporated into the confidence level, not ignored.

Intel's foundry services expansion (2020–present) Intel announced a major capital expenditure to expand foundry services to external customers, competing with TSMC and Samsung. The base rate for integrated device manufacturers successfully becoming foundry powerhouses is very low. Analysts building models assuming Intel captures 10–15% of the addressable market in foundry are implicitly forecasting above-base-rate success. This may happen, but the base rate should inform the confidence level and downside scenarios.

Starbucks' growth in China (2010–2015) In the early 2010s, Starbucks was aggressively expanding in China. The base rate for Western quick-service restaurant and coffee chains successfully capturing 5–10% market share in China is low. Analysts often modeled China as a 20–30% of growth opportunity within five years, implying above-base-rate penetration. Some of that has occurred (Starbucks has succeeded more than base rate would predict), but the base rate should have tempered the initial optimism.

Common mistakes arising from base-rate neglect​

Mistake 1: Building one deterministic model without modeling distribution of outcomes An analyst builds a point estimate (revenue will be $500M in year 5) without considering the distribution. If the base rate for this type of venture is 70% failure, the appropriate distribution should weight failure heavily, not assume the one forecast case will occur with 80%+ probability.

Mistake 2: Confusing confidence in the story with probability of success An analyst can be 95% confident that she understands the CEO's strategy and the market opportunity, while still believing only 35% probability that the strategy succeeds. These are not the same. A compelling story does not increase base-rate odds.

Mistake 3: Assuming unique factors push every case above average Every case has some unique factors. Almost every CEO believes her company is above average. The issue is not whether unique factors exist; it is whether they are substantial enough to push outcomes above the base rate. Most are not.

Mistake 4: Misidentifying the reference class An analyst covering a software company entering a new vertical may think "This company is best-in-class, so I'll use the base rate for best-in-class entries" (~40% success). But data on best-in-class entries is sparse and survivor-biased. The broader reference class (all entries into new verticals) is more reliable (~20–30%).

Mistake 5: Assuming management quality guarantees base-rate outperformance Strong management increases odds of beating the base rate, but does not guarantee it. Of the top-quartile management teams (by historical track record), 40–50% still underperform base-rate expectations in new domains. Management quality should increase your probability estimate from 20% to maybe 35–45%, not to 80%.

FAQ​

Q: What is a good source for historical base rates in equity research? A: Academic research (published in journals or by organizations like NBER), industry consultant reports (from firms like McKinsey, Bain), and proprietary databases of historical company data. Standard databases like Compustat, CapitalIQ, and Morningstar contain historical data that can be mined for base rates. Your own historical records of analyst forecasts vs outcomes is also valuable.

Q: How precise should my reference class be? A: More precise is usually better, but trade off precision against sample size. "Software company entering new vertical" is more precise than "company entering new vertical," but if only 8 data points exist in the former, use the latter. If 100+ data points exist in the more precise class, use that. Aim for at least 20–30 historical cases in your reference class.

Q: Should I use base rates as a hard ceiling on my forecast? A: No. Base rates should inform your probability assessment, not determine your forecast. If the base rate for success is 25%, your forecast might assume 40% probability (due to specific-case factors), which would imply a scenario-weighted value. But do not ignore the 60% failure scenario.

Q: How do I explain to a CEO or CFO that their growth plan is fighting base-rate odds? A: Carefully. Point out that you are not questioning the quality of the plan or the team, only noting that historical reference cases show this type of expansion is difficult. Then ask what makes this case different from the historical cases, and whether that difference is large enough to justify above-base-rate confidence.

Q: Can base rates ever be wrong? A: Yes. Base rates can become stale or unstable if the underlying business environment shifts dramatically. The base rate for cloud adoption among enterprises was 10% in 2005 and 70% in 2015. Technology change can shift base rates. But in equity research, base rates change slowly, and ignoring them for that reason is usually an error. More likely, the analyst is using an obsolete reference class.

Q: If I account for base rates, am I giving up upside capture on genuine outliers? A: No. Incorporating base rates means you assign lower probability to success, but you still model it. If a company is an outlier and succeeds, your forecast will have been pessimistic, but you will still own the position (at a lower expected value). You capture the upside; you just do not overweight it in your base case.

  • Reference class forecasting: The discipline of making predictions by identifying a historical reference class and using its distribution as your prior, then adjusting for case-specific factors.
  • Calibration in forecasts: Whether analysts' confidence levels (e.g., "75% confident") actually match historical accuracy rates; base-rate thinking improves calibration.
  • Survivor bias in analyst data: Why looking only at successful case studies (companies that succeeded in new markets) creates a biased base rate; failed attempts must be included.
  • Analyst overconfidence and base rates: Research showing that analysts are most overconfident exactly where base rates are most pessimistic.
  • Scenario analysis and base rates: How to structure scenario modeling that weights success and failure scenarios according to base-rate probabilities rather than analyst optimism.

Summary​

Base rates are powerful. In domain after domain—new product launches, executive transitions, geographic expansions, new market entries—historical frequency predicts outcomes better than specific-case analysis alone. Yet analysts routinely ignore base rates in favor of narratives, specific-case strengths, and optimistic forecasts.

The fundamental error is treating base rates as irrelevant background noise and treating the specific case as the primary signal. The correct approach is to treat the base rate as your prior and the specific-case analysis as your adjustment. If the base rate says "70% failure," and your specific-case analysis says "this management team is exceptional," your posterior might be "50% failure." Not "5% failure."

For fundamental investors, incorporating base rates into forecasts means accepting that most ambitious plans—new markets, new products, new verticals—fail. Success is possible, and outliers do exist. But the default assumption should be that the case looks more like the median case in its reference class than like the outlier. Then examine what would need to be true for this case to beat the base rate, and whether you have evidence for those conditions.

Next​

Read on to Building models with too many precise assumptions to explore how analysts construct models with false precision, compounding assumption errors rather than bounding them.

Word count: 2,104 | Keyword density (ignoring base rates, base rate, base rates, reference class): 1.4%