Part I: A Confession That Took Forty Years
Though I began my professional career as an attorney, I have been studying, practicing, and teaching finance for a very long time. It has been long enough to appreciate that the field has matured from a relatively young discipline struggling to establish intellectual respectability into a sophisticated edifice of mathematical models, Nobel Prize-winning theories, and professional certifications that have become the lingua franca of business schools, investment banks, and corporate finance departments around the world. And throughout most of that history — almost from the very beginning — I have been perplexed and troubled by something.
That something is "risk."
Not risk in general. Risk as it is defined, measured, and used in the standard finance curriculum. More specifically, the idea — embedded in virtually every introductory and intermediate finance course taught in business schools across the world — that risk can and should be measured by the variance or standard deviation of returns. That volatility is risk. That beta, derived from that same statistical apparatus, is the single number that should govern how we price capital, evaluate investments, and make decisions about the allocation of resources.
I want to be clear about what troubles me, because it is easy to misread this kind of critique as the grumbling of a practitioner who never liked the math (from the very beginning I was fascinated by the math), or a contrarian who enjoys poking at established ideas for sport (I am not smart enough for that). It is neither of those things. What troubles me is something more fundamental: that the field largely knows the critique — it has been articulated, documented, and published in the most prestigious journals — and yet the standard curriculum carries on as though that critique did not exist, or existed only as a footnote. Students are taught a framework with great confidence, the weaknesses of that framework are rarely confronted head on, but rather relegated to one or more footnotes, and an entire generation of financial professionals goes out into the world equipped with tools whose conceptual foundations are considerably shakier than they were led to believe.
I do not write this as an academic paper — I have no PhD and am not qualified to do so. I do not pretend to offer original research. What I am going to try to do instead is something I have wanted to do for a very long time: trace the argument honestly, from foundations to consequences, in a way that is accessible to someone with a serious undergraduate background in finance but does not require a doctoral degree. In previous writings I have poked at this; but never with the intent to provide a comprehensive review. If this article is useful to a student who senses that something is off but cannot yet articulate what it is, I will consider it a huge success.
Part II: What We Are Taught — And Why It Seems Reasonable
Let me start by being fair to the standard framework, because it certainly has earned and deserves that. The mean-variance approach to risk — the intellectual core of Modern Portfolio Theory as developed by Harry Markowitz in his landmark 1952 paper[1] — was a genuine and important intellectual achievement. Before Markowitz, portfolio management was largely intuitive. Diversification was understood in a vague, folk-wisdom sort of way ("don't put all your eggs in one basket") but had never been given rigorous mathematical form. What Markowitz did was formalize the relationship between risk and return at the portfolio level, show mathematically that diversification reduces risk, and derive the concept of an "efficient" portfolio — one that maximizes expected return for a given level of risk.
That was a real contribution. So was the Capital Asset Pricing Model (CAPM), developed independently by William Sharpe, John Lintner, and Jan Mossin in the 1960s.[2] CAPM took Markowitz's portfolio theory and added a market equilibrium argument to derive a strikingly clean result: in equilibrium, the only risk that should be rewarded is systematic risk — the risk that cannot be diversified away — and that risk is captured by a single number, beta. An asset's expected return should be a linear function of its beta. Everything else cancels out in a diversified portfolio.
The appeal of this framework is obvious. It is simple and elegant. It is mathematically coherent. It tells a clear story: bear more systematic risk and you will be compensated with higher expected return. And it generates numbers — betas, expected returns, costs of equity — that can be used in practical applications from portfolio construction to corporate valuation to capital budgeting. The entire apparatus of discounted cash flow analysis, the weighted average cost of capital, the Security Market Line — all of it rests on this foundation. It is not difficult to see how an entire professional ecosystem grew up around it.
Yet the moment you press on almost any part of this framework, things start to give way. And not merely at the edges. At the foundation.
Part III: The Theoretical Critique — Where the Foundation Cracks
What Markowitz Actually Assumed — And Didn't
Mean-variance optimization is only theoretically valid under one of two conditions. Either asset returns must be normally distributed, or investors must evaluate every investment solely on the basis of its expected return and variance — caring about nothing else in their decision-making, no matter what the distribution of outcomes looks like. These are not simplifying assumptions that get relaxed later. They are load-bearing conditions. If neither holds — and it appears at least to me that neither does — then variance is not a complete or sufficient statistic for risk, and everything built on top of it is constructed on uncertain ground.
Let me take the normal distribution assumption first. As I have pointed out in several other pieces and is universally understood, the normal distribution has thin tails. In a normally distributed world, extreme events are astronomically rare — the probability of a six-standard-deviation event is negligible to the point of being effectively zero. But anyone who was paying attention in 1987, in 1998 (the collapse of Long-Term Capital Management), in 2001, or in 2008 (or has studied these events) knows that financial markets do not behave this way. Events that should be essentially impossible under a normal distribution happen with disturbing regularity. This is not a matter of bad luck or unusual circumstances. It is a structural feature of how financial markets work.
The mathematician Benoit Mandelbrot argued this forcefully beginning in the 1960s.[3] Mandelbrot was certainly not a finance person. He was a towering figure in mathematics — the father of fractal geometry — and he brought fresh eyes to the problem of financial return distributions. What he saw was that actual return distributions had fat tails: extreme events occurred far more often than the normal distribution predicted. He proposed that stable Paretian distributions were a better fit. As best I can tell, the finance establishment largely ignored him, not because he was wrong, but because his alternative was mathematically inconvenient. Stable Paretian distributions can imply that variance is undefined — literally infinite. If that is true, then measuring standard deviation is not just imprecise. It is, in a technical sense, meaningless.
The second condition — that investors evaluate investments solely on expected return and variance — is, if anything, even more problematic.[4] To see why, consider what this actually requires: an investor who is genuinely indifferent between two investments with the same mean and variance, regardless of how differently their outcomes are distributed. An investment that occasionally produces catastrophic losses but has the same average return and the same variance as a steadier one would be treated as identical. No real investor behaves this way — and the reason the assumption was adopted had nothing to do with its realism. It makes the optimization problem mathematically tractable. It is a convenience, not an insight.
Markowitz himself was aware of these problems. He explicitly wrote[5] that semi-variance — measuring dispersion only on the downside — was theoretically superior to variance as a measure of risk. He set it aside not because he thought variance was better, but because the computational burden in 1952 would have been prohibitive. The field built an entire architecture on a measure its own founder acknowledged was flawed and second-best.
The Symmetry Problem — Upside and Downside Are Not the Same Thing
At the most intuitive level, variance punishes upside deviations exactly as much as downside ones. If an asset returns 30% when you expected 10%, that positive surprise contributes to measured variance exactly as much as a return of -10% would. But no investor experiences these outcomes as equivalent. Gaining more than expected is not "risky" in any meaningful sense. It is good. The risk investors actually worry about is the risk of losing money, or of falling short of some minimum threshold they need to meet.
It is worth noting that a framework addressing exactly this concern was published in the same year as Markowitz's seminal paper. Andrew Roy's "Safety First" criterion, published in 1952,[6] proposed that investors should focus on minimizing the probability of falling below a critical threshold — a target return or minimum acceptable outcome. Roy's framework is, in many ways, more intuitively aligned with how real investors think about risk. It largely never appeared in the finance curriculum. Markowitz's work, powered by the elegance of its mathematics, became the foundation of everything that followed.
CAPM: Compounding the Problems
CAPM is built on two layers of assumptions. The first layer is inherited from mean-variance optimization: either returns are normally distributed, or investors evaluate investments solely on expected return and variance. The second layer is unique to CAPM's market equilibrium argument: all investors share identical expectations about future returns; a risk-free asset exists and everyone can borrow and lend at the same rate; there are no taxes, transaction costs, or constraints; and all assets are tradeable and infinitely divisible. Both layers must hold simultaneously for beta to be the single relevant measure of risk. Both are problematic.
The empirical record is damning. The anomalies literature in academic finance — now five decades old — is essentially a sustained documentation of CAPM's predictive failures. Value stocks have systematically outperformed growth stocks in ways CAPM cannot explain. Small-cap stocks have outperformed large-cap stocks. Momentum has been robust across markets and time periods. And perhaps most embarrassingly for the theory, low-volatility stocks have tended to outperform high-volatility stocks on a risk-adjusted basis. This last anomaly — the "low-volatility puzzle"[7] — is a direct contradiction of CAPM's central prediction.
The field's response to the anomalies was to add factors. Eugene Fama and Kenneth French documented in a landmark 1992 paper[8] that beta alone did not explain differences in returns across individual stocks. They subsequently proposed a three-factor model, then a five-factor model. Mark Carhart[9] added momentum. The proliferation of factors has continued to the point where somewhere between a few dozen and a few hundred "factors" have been proposed in the academic literature.[10]
This should prompt some uncomfortable questions. If the single-factor model failed empirically and we responded by adding factors to fit the data, what is the theoretical foundation for the multi-factor models? The honest answer is that it is unclear. Unlike CAPM, which at least had a coherent equilibrium story, the additional factors were identified empirically, and the theoretical explanations have been disputed ever since.
The Problem Nobody Talks About: Risk as a Property of the Investor, Not the Asset
There is a deeper problem embedded in the standard framework that is rarely surfaced in textbooks. The framework treats risk as an objective property of an asset — the variance of its returns — as though risk were something that attached to the asset the way mass attaches to a physical object.
But this is unquestionably a dubious assumption. The same Treasury bond is "risk-free" for a buy-and-hold investor who holds it to maturity but presents meaningful interest rate risk for a financial institution with short-duration liabilities that marks it to market daily. A volatile stock in a well-understood business, purchased at a significant discount to what a careful analyst believes it to be worth, may present far less genuine risk to a patient, knowledgeable investor than a stable, low-beta stock in a business whose competitive position is quietly deteriorating.
Risk, in other words, is at least partly investor specific. It depends on who is holding the asset, under what constraints, over what time horizon, with what existing portfolio, and with what level of understanding of what they actually own.
Frank Knight made a related but deeper distinction back in 1921[11] that deserves far more attention than it typically receives in finance education. Knight distinguished between risk — situations where outcomes are uncertain but can be assigned probabilities — and uncertainty, where the structure of possible outcomes is itself unknown. An equity investor trying to assess the prospects of a company in a rapidly changing industry, against competitors who may not yet exist, in a regulatory environment that may shift — that investor is operating largely in a world of Knight's uncertainty. Applying variance mathematics to uncertainty is not just imprecise. It may be, as Knight would have said, a category error.
Part IV: Why the Field Persists — The Sociology of a Flawed Consensus
The Teaching Problem
Finance curricula are built to be coherent and sequential. Markowitz leads to CAPM; that leads to cost of equity; in turn that leads to discounted cash flow analysis, valuation and to capital budgeting. Each piece connects cleanly to the next. If you introduce the full critique at the first step — "these distributional assumptions are empirically wrong, the equilibrium conditions are unrealistic, and the measure of risk is theoretically problematic" — you have undermined everything that follows before students have had a chance to learn it. Pedagogically, that would almost certainly be impossibly difficult to manage. So, the critique gets deferred, softened, mentioned in footnotes, or left to advanced courses that most students never take.
Institutional Inertia and Professional Embedding
The standard framework is not just taught — it is deeply embedded in professional practice. The CFA curriculum is organized around it. Regulatory capital requirements in banking incorporate it. Investment policy statements at major institutions reference Sharpe ratios and tracking error. Performance attribution systems are built on it. The infrastructure of the financial industry has been constructed on this foundation over decades. The cost of replacing it — even if a clearly superior alternative were available — would be enormous.
The Sociology of a Discipline
This may be the most underappreciated factor. Academic finance built its identity as a rigorous quantitative discipline — the idea that finance could be as precise as physics, that markets could be modeled with the same tools used to describe the motion of particles.[12] This identity is valuable. It justifies the discipline's place in research universities, supports its claim to scientific credibility, and distinguishes academic finance from mere business judgment or market commentary. Challenging the mathematical foundations is not merely an intellectual disagreement. It threatens the field's professional identity.
Practitioners who operate outside the standard framework — Warren Buffett being the most obvious example — are not taken as evidence that the framework is wrong. They are treated as anomalies to be explained away.[13] Mandelbrot's experience is instructive here. He was a mathematician of the first rank, the founder of an entirely new branch of mathematics, a professor at Yale, a Fellow of virtually every major scientific society. He spent decades arguing — with detailed empirical evidence — that financial return distributions had fat tails and that the standard framework's distributional assumptions were fundamentally wrong. The finance establishment acknowledged his work, published it in its journals, and largely went on doing exactly what it had always done.
The Alternatives Are Harder
There is also a simpler explanation: the standard framework is teachable. The alternatives are considerably harder. Extreme Value Theory requires serious statistical background. CVaR optimization is computationally demanding. Semi-variance breaks the clean matrix algebra that makes portfolio optimization tractable. And the permanent capital impairment framework requires deep business judgment that resists reduction to formula. If you have fourteen weeks and students with heterogeneous quantitative backgrounds, you teach what is teachable. The alternative is mentioned in a footnote.
Part V: Practical Alternatives — Thinking About Risk More Honestly
Fixing the Symmetry Problem: Semi-Variance and the Sortino Ratio
Semi-variance measures dispersion only below the mean or below some target return. It is the measure Markowitz himself identified as theoretically superior. The Sortino ratio[14] operationalizes this by replacing the standard deviation in the denominator of the Sharpe ratio with downside deviation, giving you a measure of return per unit of the risk that actually matters: the risk of outcomes worse than your target.
Why did semi-variance never fully displace variance? Not because variance is better. Semi-variance is not additive across assets the way variance is.[15] The field kept the inferior measure because it was easier to aggregate. That is a significant confession, and it is almost never made explicit in the classroom.
Fixing the Tail Blindness Problem: VaR, CVaR, and Expected Shortfall
Value at Risk emerged from risk management practice — particularly from JP Morgan in the 1990s — as a way to answer a question that standard deviation does not address: what is the maximum loss over a given period at a given confidence level? But VaR has a critical flaw: it tells you nothing about what happens in the worst cases beyond the threshold. If two portfolios both show a 99th percentile VaR of $10 million, you know they each lose less than $10 million on 99 out of 100 days. What you do not know is what happens on the remaining one day in a hundred — Portfolio A might lose $12 million on its worst days while Portfolio B loses $80 million. Worse, VaR is not subadditive[16] — combining two portfolios can produce a VaR higher than the sum of their individual VaRs, which violates a basic logical requirement for a coherent risk measure.
Conditional Value at Risk — CVaR, also called Expected Shortfall — fixes VaR's most serious problem. Instead of asking what the loss is at the 99th percentile threshold, CVaR asks: given that you have had one of those bad days when losses exceed the threshold, what is the expected loss on average across all such outcomes? It looks at the entire tail beyond the threshold, not just its boundary. CVaR is mathematically coherent, is subadditive, and is far more sensitive to tail risk than VaR. The Basel III regulatory framework eventually incorporated Expected Shortfall as a supplement to VaR — which is essentially regulators acknowledging, after 2008, that VaR had failed them.
Fixing the Distribution Problem: Extreme Value Theory
Rather than assuming a normal distribution and working across the full return distribution, Extreme Value Theory[17] focuses specifically on the statistical behavior of extremes. It has solid theoretical foundations — limit theorems for tail behavior that are analogous to the Central Limit Theorem for averages — and it does not require normality. EVT is used in catastrophe insurance, flood modeling, and increasingly in financial risk management. It is technically demanding and requires large datasets to estimate tail parameters reliably, but it is honest about the problem in a way that standard deviation is not.
Abandoning the Single Number: Scenario Analysis and Stress Testing
Rather than compressing risk into a single number, scenario analysis asks: what happens to this portfolio under specific conditions? A 2008-style credit crisis. A 1970s inflationary environment. A deflationary shock. A rapid rise in interest rates. This approach is intellectually more honest than any single-number risk measure because it does not pretend to summarize an uncertain future in one statistic. It forces explicit thinking about what environments would be damaging and why. The limitation is that you can only stress-test what you imagine. The risks that actually materialize are often the ones that were not in the scenario set.
The Permanent Capital Impairment Framework
The most radical alternative is also the oldest. The value investing tradition — rooted in Benjamin Graham's work in the 1930s[18] and most prominently embodied in the practice of Warren Buffett — rejects volatility as the relevant measure of risk almost entirely. In this framework, risk is the probability of permanent capital impairment: the probability that you will not recover your capital in any reasonable time horizon.
Under this definition, a high-volatility stock in a business you understand deeply, purchased at a significant discount to its estimated intrinsic value, can be genuinely low risk. A stable, low-beta stock in a business whose competitive position is quietly deteriorating can be genuinely high risk. The analytical work shifts entirely: you are no longer estimating return distributions but assessing business quality, competitive durability, balance sheet resilience, and — crucially — margin of safety.
Margin of safety deserves particular emphasis because it is itself a risk management tool, and an intellectually honest one. If your estimate of a business's intrinsic value is uncertain — and it always is — then purchasing at a substantial discount to that estimate means that your estimate can be wrong by a meaningful margin and you still do not lose money. The discount absorbs the uncertainty. This is not quantitatively sophisticated in the way that CVaR optimization is sophisticated. But it may be more practically effective than false precision in inputs that are themselves deeply uncertain.
Part VI: Where the False Precision Has Real Consequences — Cost of Equity and Valuation
Everything described so far might seem like an abstract theoretical debate. It is not. The theoretical failures transmit directly into practice through the cost of equity, and the consequences are demonstrably real.
The Cost of Equity Formula and Its Fragile Inputs
The CAPM formula for the cost of equity is: Re = Rf + β(Rm − Rf). Three inputs. Each one carries its own serious problems, and they compound.
Start with the risk-free rate. This looks like a clean input — just use the current 10-year Treasury yield. But risk-free to whom, for what purpose, over what horizon? The cost of equity swings dramatically based on the interest rate environment prevailing when the calculation happens to be performed, which is uncomfortable if you believe you are measuring something fundamental about the business being valued.
Beta is worse. It is estimated from historical price data — typically two to five years of weekly or monthly returns regressed against a market index. The estimate is noisy. Standard errors are large. Two analysts using different time periods or return frequencies will often produce meaningfully different betas for the same company. Beta is also unstable over time: it changes as a company's capital structure changes, as its business mix evolves, as market conditions shift. And most damaging of all, as Fama and French documented,[19] the empirical relationship between beta and realized returns is weak.
The equity risk premium — the term (Rm − Rf) — is where the imprecision becomes most troubling. Historical estimates for US equities range roughly from 3% to 8% depending on the period chosen, the market index used, and whether you use arithmetic or geometric means. That is not a measurement error. That is a range wide enough to render any resulting cost of equity calculation nearly meaningless as a precise number. Dimson, Marsh, and Staunton's long-run work across multiple countries[20] suggests the historical premium may overstate the forward-looking premium due to survivorship bias — we are studying markets that survived. The practical result is that two disciplined, serious analysts can produce cost of equity estimates for the same company that differ by three or four hundred basis points using completely defensible methodologies. Run those through a discounted cash flow model and you get valuations that differ by 30 to 50 percent or more.
The Compounding Problem in DCF Analysis
The mathematics of discounting amplify every input error. A cash flow ten years out discounted at 8% versus 10% produces values that differ by roughly 21 percent. At twenty years the same comparison produces values differing by roughly 38 percent. Terminal value — which in most DCF analyses represents somewhere between 60 and 80 percent of total estimated value — is discounted at the full rate over the full horizon and divided by (r − g), meaning that errors in the discount rate and in the perpetual growth rate interact multiplicatively. Small changes in either input produce enormous swings in estimated terminal value. The terminal value calculation is where the false precision does its greatest damage. And the result is then reported to two decimal places.
A Circularity Problem Clearly Understood but Rarely Discussed
There is a deeper problem embedded in using market-derived inputs to value market-traded assets. Beta is estimated from market prices. The equity risk premium is estimated from market returns. Both inputs therefore reflect whatever the market currently believes about risk and return. You then use those inputs to evaluate whether the market is pricing something correctly. But if the market is mispricing the asset — which is precisely what you are trying to determine — your inputs already embed that mispricing. The tool is partially self-referential. It tends to confirm market prices rather than challenge them, which undermines the very purpose of the exercise.
What More Honest Practice Looks Like
Several more intellectually honest approaches exist and are used by careful practitioners. The build-up method[21] constructs the cost of equity additively rather than through a single formula: start with the risk-free rate, add an equity risk premium, then add specific risk premiums for company size, business-specific risk, illiquidity, leverage, and industry characteristics. Each component can be examined and debated on its merits. The judgment calls are visible rather than buried inside beta.
Scenario-based valuation acknowledges uncertainty honestly by producing a range of values rather than a single number. The reverse DCF,[22] championed by Damodaran among others, is perhaps the most underused and intellectually honest approach available. Rather than assuming inputs and calculating value, it starts from the current market price and asks: what growth rate and discount rate assumptions are implied by this price? The analyst then evaluates whether those implied assumptions are realistic. Using multiple valuation methodologies simultaneously and treating their convergence or divergence as information is also more honest than relying on a single model.
Part VII: Synthesis — What a Thoughtful Student or Practitioner Should Do
I want to be careful here not to leave the reader in a state of despair or even discouragement. The conclusion is not that risk cannot be analyzed, that valuation is hopeless, or that the standard framework should simply be discarded. The conclusion is more nuanced and, at least in my opinion, more useful than that.
The standard framework is a coherent system with real intellectual content. Learning it is not wasted time. Understanding mean-variance optimization, CAPM, beta, and discounted cash flow analysis is necessary precisely because these are the tools the professional world uses, the language in which capital markets communicate, the framework within which countless important decisions are made.
But there is an enormous difference between understanding a framework and believing it is theoretically sound and practically correct. The practitioners who navigate financial markets most effectively are those who use the standard framework as a discipline for organizing thinking rather than as a measurement tool that produces authoritative answers. They run the DCF. They calculate the cost of equity. And then they ask: what would have to be true for this price to make sense, and do I actually believe those things? The model is a starting point for judgment, not a replacement for it.
What the alternatives share — semi-variance, CVaR, scenario analysis, the permanent impairment framework — is that they require you to make explicit choices that the standard framework buries in assumptions. What are you afraid of? What is your time horizon? What scenarios do you think are plausible? What is the probability that you have the business fundamentally wrong? These are uncomfortable questions. They resist reduction to formula. They require judgment. And they are, in my opinion, ultimately, the right questions.
Part VIII: Concluding Thoughts — The Privilege and Responsibility of Discomfort
I said at the beginning that I have been troubled by the standard treatment of risk for most of my professional life in finance. Having finally put the argument together in one place, I want to say something about what that discomfort means and why I think it matters.
Finance is not physics. It is not engineering. It is a discipline that studies human decision-making under uncertainty, in markets made up of human beings with limited knowledge, shifting beliefs, and institutional constraints that change over time. The aspiration to reduce this to a mathematical science as clean and precise as Newtonian mechanics was understandable — and the attempt produced genuine insights. But the aspiration was always partly a conceit, and the cost of pursuing it too faithfully has been a curriculum that systematically overstates what it knows and understates the uncertainty that pervades its most important conclusions.
Students who feel uncomfortable with this — who sense that something does not quite add up, who wonder why the models seem cleaner than the reality they are supposed to describe — are not missing something. They are seeing something. The discomfort is an asset, not a deficiency. It is the beginning of the kind of critical thinking that distinguishes a careful student or practitioner from someone who runs the model and reports the output.
The framework is a tool. Tools are useful when you understand what they are designed to do, what they can and cannot measure, and where their output should be trusted and where it should be challenged. A tape measure is a good tool for measuring the length of a table. It is not a good tool for measuring temperature. Using it for the wrong purpose does not make the tape measure bad — it makes the user careless. You should think of the standard risk framework as a tape measure; but always remember that financial markets are driven not only by distance but by a great deal of temperature.
The best practitioners — in investment management, in corporate finance, in risk management — undoubtedly know this. They use the framework while maintaining a healthy skepticism about its outputs. They run the numbers and then ask whether the numbers make sense. They hold the model's conclusions loosely and the underlying business judgment firmly. They are aware that the confidence the numbers project often exceeds the confidence the analysis warrants.
That is not a message that the academic curriculum delivers often enough or forcefully enough. It should.
I hope this article is useful to someone who is willing to sit with the discomfort a little longer than the curriculum requires — and to think carefully about what they actually know, what they are assuming, and what remains genuinely uncertain. In finance, the difference between those three categories matters enormously. Getting them confused is not just an intellectual error. It can have meaningful and sometimes disastrous consequences.