There is something psychologically powerful about a number with several digits. A business case produces an NPV of $438,721. A property valuation comes back at $2,743,500. An investment model estimates an internal rate of return of 17.84%. A retirement calculator says that an investor will have $1,284,613 at age 65. The numbers look different from opinions. They look measured, processed and objective.

That appearance can be dangerously misleading. The calculation may indeed be mathematically flawless. The spreadsheet may contain no broken formulas. Every cell may reference the correct input. The discounting may be performed correctly, the cash flows may be summed correctly, and the NPV formula may be implemented exactly as intended.

And yet the answer may still be profoundly wrong.

The reason is simple: financial models do not calculate the future; they calculate the consequences of assumptions about the future.

This distinction is easy to forget because a model converts uncertain judgments into highly precise numerical outputs. A manager may estimate that revenue will grow by 8%, margins will reach 22%, capital expenditure will remain at a particular percentage of sales and the business will eventually grow at 3%. Once those assumptions enter a spreadsheet, they become rows, columns and formulas. The uncertainty that existed in the original judgment becomes visually invisible.

That is the beginning of false precision.

The Office of the Comptroller of the Currency’s current model-risk guidance makes essentially the same fundamental point from a regulatory perspective: models are simplified representations of real-world relationships, and assumptions, data limitations and simplifications create uncertainty in model outputs.

The danger, therefore, is not mathematics. The danger is confusing mathematical exactness with knowledge.

A Model Is Not the Future

A financial model is an abstraction. It takes a complicated reality and deliberately leaves things out.

That is not a weakness in itself. Without simplification, meaningful analysis would often be impossible. A model is useful precisely because it strips reality down to a manageable structure.

The problem begins when the simplification is forgotten.

Consider a basic discounted cash-flow model:

$NPV = \sum_{t=1}^{n}\frac{CF_t}{(1+r)^t}$

The equation is not uncertain. Once the cash flows CFt , discount rate r, and time periods t are specified, the mathematical result is determined.

But almost every economically important input is uncertain.

What will sales be five years from now? What will inflation be? Will margins expand or contract? Will competitors enter the market? Will interest rates remain elevated? Will regulation change? Will customers behave differently? Will a technological substitute appear? Will the business survive?

The formula does not know. It simply accepts whatever the analyst tells it.

This is why a useful way to think about a financial model is as a conditional statement:

If these assumptions are approximately correct, this is the numerical consequence.

That is fundamentally different from saying:

This is what will happen.

Valuation specialist Aswath Damodaran has emphasized that valuation is inherently uncertain and that the precision of a valuation should not be mistaken for the quality or certainty of the underlying estimate.

The model is therefore not an oracle. It is an argument written in mathematics. And, like any argument, its conclusion is only as strong as its premises.

Where the Assumptions Hide

One of the most dangerous characteristics of financial models is that assumptions rarely announce themselves as assumptions.

A number such as 8.0% revenue growth looks like a fact once it sits inside a spreadsheet. It is not a fact. It is a belief about the future. The same applies to assumptions concerning operating margins, customer acquisition, inflation, wages, tax rates, capital expenditure, working capital, default rates, asset prices, interest rates and countless other variables.

The model converts all of them into numerical inputs, creating an illusion that they possess the same epistemic status. They do not. Historical revenue may be observable. Next year’s revenue is estimated. Revenue ten years from now is largely a structured conjecture.

The hidden hierarchy of model inputs

Input	Typical evidence base	Degree of uncertainty
Historical revenue	Observed financial data	Relatively low
Current interest rate	Observable market data	Relatively low
Near-term contractual cash flow	Contractual information	Low to moderate
Next-year revenue	Forecast	Moderate
Five-year margin	Strategic assumption	High
Ten-year market share	Competitive forecast	Very high
Perpetual growth rate	Long-term assumption	Extremely high
Terminal value	Derived from distant assumptions	Potentially extreme

The problem becomes particularly serious when analysts build long chains of assumptions. Suppose a model assumes that a company’s market share rises from 8% to 14%, revenue grows at 10%, operating margins increase from 12% to 20%, capital intensity declines and the company ultimately grows forever at 3%. None of those assumptions is necessarily absurd.

The problem is that they must all be approximately right at the same time for the resulting valuation to deserve the confidence implied by a precise number such as $438,721. A model can therefore contain individually reasonable assumptions that collectively create an unreasonable conclusion.

Forecasting Error: The Longer the Horizon, the Greater the Problem

Forecasting is not simply difficult because the future is unknown. It becomes progressively harder because errors accumulate.

A forecast for next month is constrained by current conditions. A forecast for ten years from now must incorporate multiple layers of uncertainty: economic conditions, technological change, competition, regulation, consumer behaviour, financing conditions and the company’s own strategic responses.

Research on analyst forecasts provides an important empirical warning. A 2021 NBER study found that analyst forecasts performed better than statistical forecasts over short horizons but underperformed them at longer horizons, while both forecast noise and bias increased with the forecasting horizon.

That finding has an uncomfortable implication for long-range financial models. The distant years may occupy only a few columns in a spreadsheet, but they can contain enormous uncertainty.

Illustrative forecast-error expansion

Forecast horizon	Example forecast	Potential confidence
1 year	Revenue = $110m	Relatively defensible
3 years	Revenue = $135m	Meaningful uncertainty
5 years	Revenue = $165m	High uncertainty
10 years	Revenue = $260m	Very high uncertainty
20 years	Revenue = $670m	Extremely assumption-dependent

The numbers above are illustrative rather than empirical forecasts. Their purpose is to demonstrate a structural problem: forecast horizons do not merely add more years; they add more opportunities for assumptions to diverge from reality.

A particularly dangerous practice is to extrapolate recent growth far into the future.

A company that grew 25% last year may not grow 25% for the next decade. A housing market that appreciated rapidly during one period does not automatically justify assuming comparable appreciation indefinitely. A technology company benefiting from an emerging trend may eventually encounter saturation, competition or technological displacement. Historical performance is evidence. It is not a contract with the future.

The Discount Rate: One Percentage Point Can Change Everything

The discount rate often appears to be a technical parameter. It is anything but.

In a discounted-cash-flow model, future cash flows are converted into present values by discounting them:

$PV_t=\frac{CF_t}{(1+r)^t}$

As r increases, distant cash flows become less valuable today.

As r decreases, distant cash flows become more valuable.

This creates an uncomfortable asymmetry: small changes in the discount rate can have very large effects when substantial value depends on distant cash flows.

Consider a simplified perpetual-growth relationship:

$TV=\frac{CF_{n+1}}{r-g}$

where r is the discount rate and g is the perpetual growth rate. The denominator r-g is the critical point.

If the discount rate is 10% and perpetual growth is 3%, the difference is 7 percentage points. If the discount rate falls to 9% while growth remains 3%, the difference becomes 6 percentage points. That apparently small change can materially increase the calculated terminal value.

This is not a spreadsheet error.

It is the mathematics behaving exactly as designed.

The judgment problem lies in deciding whether 9% or 10% is actually justified.

Damodaran notes that the appropriate discount rate depends on the riskiness of the cash flows and that valuation requires judgments about expected cash flows, timing and discount rates.

This is why the discount rate should never be treated as a decorative number placed near the top of a spreadsheet.

It is a major assumption about how uncertain future cash flows are and what return investors require for bearing that uncertainty.

Terminal Value and the Mathematics of the Distant Future

Terminal value may be the most misunderstood component of many financial models.

A company does not conveniently stop existing at the end of year five merely because the spreadsheet does.

So analysts generally estimate the value of cash flows beyond the explicit forecast period through a terminal value.

Under the perpetual-growth approach:

$TV_n=\frac{FCF_{n+1}}{r-g}$

This equation looks harmless. It is not. The terminal value may represent a substantial share of the total valuation, meaning that today’s calculated value can depend heavily on assumptions about a business many years into the future.

Damodaran specifically warns that small changes in the stable growth rate can significantly affect terminal value, with the sensitivity becoming particularly severe as the growth rate approaches the discount rate.

Illustrative terminal-value sensitivity

Assume:

Terminal-year cash flow = $100 million
Discount rate = 9%
Perpetual growth = variable

Perpetual growth	Terminal value
1.0%	$1.25 billion
2.0%	$1.43 billion
3.0%	$1.67 billion
4.0%	$2.00 billion
5.0%	$2.50 billion

The difference between 1% and 5% perpetual growth is only four percentage points.

Yet the resulting terminal values are dramatically different.

This illustrates an important principle:

The smaller the denominator in a valuation equation, the more dangerous apparently modest assumption changes can become.

There is another conceptual problem. A perpetual growth rate is not simply a forecast for next year’s growth. It is an assumption about sustainable growth over an indefinitely long horizon.

A 5% growth rate might be plausible for a company for several years.

It is much harder to defend as an indefinitely sustainable growth rate in a mature economy.

Terminal value therefore deserves scrutiny proportional to its contribution to the valuation. If most of the estimated value comes from the terminal period, the model is effectively making a very large bet on a relatively small set of long-term assumptions.

Sensitivity Analysis: Finding What Actually Matters

A single forecast answers one question:

What happens if everything goes according to this particular set of assumptions?

Sensitivity analysis asks a much better question:

What happens if the assumptions are wrong?

This distinction is fundamental. Suppose a project produces an NPV of $438,721 under the base case. That number alone tells us very little about the robustness of the decision. We should ask how the NPV changes when the major assumptions change.

Illustrative NPV sensitivity matrix

Discount rate ↓ / Growth →	2%	3%	4%	5%
8%	$510k	$590k	$690k	$830k
9%	$440k	$500k	$580k	$680k
10%	$380k	$430k	$500k	$580k
11%	$330k	$370k	$420k	$490k
12%	$290k	$320k	$360k	$410k

Illustrative values only; they are intended to demonstrate sensitivity rather than represent a specific investment.

The important information is no longer the central estimate. It is the shape of the answer. If a small change in two assumptions turns a positive NPV into a negative one, the investment may be fragile even though the base case looks attractive. This is where sensitivity analysis becomes a diagnostic tool rather than merely a presentation feature.

A good sensitivity analysis should identify which assumptions have the greatest influence on the decision. These may include:

revenue growth;
operating margin;
capital expenditure;
working capital;
discount rate;
terminal growth;
default rates;
inflation;
interest rates; or
exit multiples.

The objective is not to produce an enormous table of every possible variation. The objective is to discover where the model is structurally vulnerable.

Scenario Analysis: When One Forecast Is Not Enough

Sensitivity analysis changes individual variables.

Scenario analysis changes the world around the model. That distinction matters because financial variables rarely move independently. A recession might reduce revenue, compress margins, increase defaults, raise financing pressure and simultaneously change the appropriate discount rate. Changing only revenue while leaving everything else untouched may therefore underestimate the real risk.

A more realistic scenario framework might look like this:

Variable	Optimistic	Base	Adverse
Revenue growth	12%	8%	2%
Operating margin	24%	20%	14%
Inflation	2%	3%	6%
Discount rate	8%	10%	13%
Terminal growth	4%	3%	1%
NPV	High	Moderate	Negative

Again, the numbers are illustrative. The important idea is that scenarios should represent coherent states of the world, rather than arbitrary collections of optimistic and pessimistic numbers.

A useful scenario might be “persistent high inflation”, “recession”, “rapid technological disruption”, “regulatory restriction” or “strong demand expansion”. Each scenario can then alter multiple variables consistently.

This approach acknowledges something that conventional spreadsheets often hide: the future does not change one cell at a time.

The IMF has increasingly emphasized approaches that combine forecasting with scenario analysis and explicitly account for uncertainty, including nonlinearities, skewed distributions and incomplete scenario sets.

Scenario analysis is therefore not an admission that the model has failed. It is an admission that reality contains more than one plausible future.

Model Risk: When the Model Becomes the Risk

There is a peculiar paradox in quantitative finance. Models are built to manage risk. Yet models themselves create risk. This is known as model risk.

A model can produce a wrong result because its mathematics are incorrect, its assumptions are inappropriate, its data are poor, its implementation contains errors, or the model is used outside the environment for which it was designed.

The current U.S. interagency model-risk guidance explicitly identifies assumptions, model complexity, input quality and data constraints as contributors to inherent model risk. It also emphasizes that even a fundamentally sound model can exhibit high model risk when it is misapplied or misused.

That last point is especially important. Imagine a model that was developed for normal market conditions. A financial institution then uses it during a period of extraordinary market stress.

The formulas have not changed. The spreadsheet still works. But the relationship between the variables may have changed. The model is now being asked a question it was never designed to answer. That is not a mathematical failure. It is a model-use failure.

Model risk can therefore arise at several levels:

Source of risk	What can go wrong?
Data risk	Inputs are incomplete, outdated or biased
Assumption risk	Key assumptions are unrealistic
Specification risk	Important relationships are omitted
Parameter risk	Estimated coefficients or rates are wrong
Implementation risk	The model is coded incorrectly
Validation risk	Weaknesses are not detected
Use risk	The model is applied outside its intended purpose
Governance risk	Decision-makers place excessive confidence in the output

This is why model validation is not simply a technical exercise. It is also a challenge to the model’s underlying story.

Behavioural Bias: The Human Being Inside the Spreadsheet

It is tempting to imagine that spreadsheets eliminate human bias. They do not. They can simply hide it. A spreadsheet does not choose its assumptions. People do. And people bring incentives, expectations, fears and psychological shortcuts into the modelling process.

CFA Institute identifies several behavioural influences relevant to financial forecasting, including anchoring, confirmation bias, overconfidence, status-quo bias, availability bias and other psychological distortions.

Consider an analyst who believes that a company is undervalued. They may unconsciously select:

a relatively high revenue-growth assumption;
a favourable margin trajectory;
a lower discount rate;
an optimistic terminal growth rate;
and a generous exit multiple.

None of these assumptions may be individually indefensible. Together, however, they can manufacture the desired answer. The spreadsheet then appears to validate the analyst’s original belief.

This is one of the most dangerous feedback loops in financial analysis:

belief → assumptions → model → number → apparent confirmation of belief.

Research has found evidence that analyst expectations can contain systematic optimism. NBER research has documented upward-biased earnings forecasts, particularly at longer horizons, while other research has linked optimistic long-term expectations with subsequent disappointment.

The problem is not that analysts are irrational. The deeper problem is that human judgment is part of the model whether the spreadsheet acknowledges it or not.

Uncertainty Is Not the Same as Risk

One of the most important distinctions in financial modelling is the difference between risk and uncertainty. Risk generally refers to situations where possible outcomes can be described with some meaningful probability structure.

Uncertainty is more fundamental. It exists when we do not know the relevant probabilities, do not know all possible outcomes, or cannot confidently specify the underlying distribution. A financial model may be very good at calculating risk under a known structure. It is much less capable of dealing with events outside that structure.

This matters enormously during structural breaks.

A model trained on stable relationships may struggle when those relationships change. Historical correlations can weaken. Consumer behaviour can shift. Regulation can change the economics of an industry. New technologies can make previous relationships obsolete.

The Bank for International Settlements has noted the fundamental difficulty of modelling uncertainty in financial systems, emphasizing that financial interactions involve complex social behaviour that cannot be completely captured by models.

This is why a probability distribution should not automatically be mistaken for reality.

A model saying there is a 1% probability of something does not necessarily mean that the real-world probability is 1%. It may mean that the model assigns 1% based on its assumptions.

That distinction is subtle but crucial.

When More Sophisticated Models Make the Problem Worse

Complexity often creates an illusion of reliability. A simple model with ten assumptions looks obviously uncertain. A sophisticated model with thousands of variables, probability distributions, machine-learning algorithms and Monte Carlo simulations can look scientifically authoritative.

But complexity does not automatically create accuracy. Sometimes it creates precision theatre. Suppose a Monte Carlo model produces:

Expected NPV = $438,721
5th percentile = $112,403
95th percentile = $817,294

This looks impressively rigorous. But the output still depends on the assumptions governing the distributions.

If the model underestimates the probability of extreme events, the simulation may produce a beautifully precise but misleading range. If the correlations between variables are wrong, the simulated outcomes may be structurally distorted. If an important risk is absent from the model, running one million simulations does not solve the problem.

It simply simulates the wrong world one million times.

This is a fundamental lesson:

More computation cannot compensate for missing knowledge.

The IMF’s work on forecast uncertainty illustrates why sophisticated uncertainty analysis often needs to account for non-Gaussian shocks, nonlinear models, skewness and fat tails rather than relying on convenient assumptions about normally distributed outcomes.

The issue is therefore not whether a model is simple or sophisticated.

The real question is:

Does its complexity improve its representation of reality, or merely make its assumptions harder to see?

From False Precision to Decision-Useful Precision

The answer is not to abandon financial models. That would be a mistake. Models remain extraordinarily useful. They allow people to compare alternatives, understand cash-flow structures, identify important variables, test strategies and expose financial consequences that intuition alone may miss.

The objective should instead be to replace false precision with decision-useful precision. That requires changing what the model is expected to produce.

Instead of asking only: “What is the NPV?”

ask: “Under what assumptions is the NPV positive?”

Instead of asking: “What is the expected return?”

ask: “How sensitive is the return to the assumptions that matter most?”

Instead of asking: “What is the probability of failure?”

ask: “What assumptions determine that probability, and what happens if those assumptions are wrong?”

Instead of presenting one number, a robust financial analysis should communicate the structure behind that number.

A stronger modelling framework

Question	Weak modelling practice	Strong modelling practice
Forecast	One “best estimate”	Base case plus uncertainty
Discount rate	Single unquestioned rate	Justified range and sensitivity
Terminal value	One final number	Multiple defensible approaches
Risk	Historical averages only	Stress and scenario analysis
Bias	Assumed away	Explicitly challenged
Output	Precise point estimate	Range plus decision thresholds
Validation	Formula checking	Assumption, data and outcome validation
Communication	“The model says…”	“Under these assumptions…”

This changes the role of the model. It becomes less of a prediction machine and more of a decision laboratory. That is a much healthier role.

A Practical Framework for Challenging Any Financial Model

Before accepting a model’s output, a decision-maker should be able to answer several questions.

What exactly is being assumed? If an assumption is buried inside a formula, it should be extracted and made visible.

Which assumptions matter most? A one-percentage-point change in a variable that barely affects the output is less important than a small change in a variable that transforms the decision.

How far into the future are we forecasting? The farther the forecast horizon, the greater the need for humility.

How much of the valuation depends on terminal assumptions? If terminal value dominates the result, the headline valuation deserves particular caution.

What happens under adverse but plausible conditions? A model that only works in the base case is not necessarily a robust decision model.

Are the variables realistically correlated? Economic shocks rarely affect only one variable.

What historical evidence supports the assumptions? A forecast should not simply be an extrapolation of optimism.

What would prove the model wrong? A model that cannot be falsified becomes difficult to challenge.

Who benefits from the assumptions? Incentives matter. Forecasts created for financing, fundraising, acquisitions or internal approval may face pressures that influence their inputs.

When was the model last challenged? A model is not permanently valid simply because it once worked.

These questions transform modelling from spreadsheet production into analytical discipline.

The Deeper Problem: Numbers Can Hide Uncertainty Better Than Words

This may be the most important point. People are often suspicious of vague language e.g.

“Strong growth is expected.”

“Returns should remain attractive.”

“The investment appears promising.”

These statements sound uncertain.

Replace them with: Expected IRR = 18.37%

and suddenly the uncertainty seems to disappear. But it has not disappeared. It has simply been compressed into assumptions. This is why numbers can sometimes be more misleading than words. A vague statement openly signals uncertainty. A highly precise number can conceal it.

The psychological effect is powerful because humans often interpret numerical precision as evidence of analytical quality. But the number of decimal places says nothing about the quality of the forecast.

An IRR of 17.84% is not necessarily more credible than an IRR of 18%. If the underlying cash flows are uncertain, the extra two decimal places may be meaningless. The same principle applies to valuations. A valuation of $438,721 may suggest that the analyst knows the value to within a dollar.

In reality, the economically meaningful conclusion might be closer to:

“The estimated value is somewhere within a broad range, and the investment becomes unattractive if revenue growth falls below X or the discount rate rises above Y.”

That answer contains less numerical drama. It may contain much more useful information.

What a Better Financial Model Should Look Like

A robust financial model should not merely output a number. It should reveal the conditions under which the number remains credible. The model should distinguish observed data from assumptions, short-term forecasts from long-term forecasts, controllable variables from uncontrollable variables, and ordinary risk from genuinely uncertain outcomes. It should make sensitivity visible rather than burying it. It should use scenarios to explore coherent alternative futures. It should test whether conclusions survive reasonable changes in assumptions. It should identify which variables have the greatest influence on the decision.

And, critically, it should make it difficult for the user to forget that the future remains outside the spreadsheet.

Financial institutions increasingly treat model development, validation, monitoring and governance as distinct elements of model-risk management precisely because simply checking whether a formula runs correctly is not enough.

The same principle should apply much more broadly.

A financial model should be challenged before it is trusted.

Research Evidence at a Glance

Issue	What research and professional guidance indicate	Practical implication
Model assumptions	Models simplify real-world relationships and depend on assumptions, data and hypotheses	Assumptions must be challenged, not merely documented
Forecasting	Forecast noise and bias can increase with longer horizons	Long-term forecasts deserve wider uncertainty ranges
Discount rates	Valuation depends critically on cash flows, timing and discount rates	Small rate changes can materially alter long-duration valuations
Terminal value	Small changes in perpetual growth can have large valuation effects	Terminal assumptions require particular scrutiny
Scenario analysis	Alternative scenarios can complement statistical forecasts	Decision-makers should examine multiple plausible futures
Model risk	Models can fail through poor assumptions, inputs, specification or misuse	Model validation must go beyond checking formulas
Behavioural bias	Analysts and decision-makers can exhibit systematic optimism and other biases	Independent challenge is essential
Uncertainty	Not all future outcomes can be represented adequately by historical distributions	Stress testing and judgement remain necessary

Conclusion: The Most Dangerous Number Is the One That Looks Certain

Financial models are not dangerous because they use mathematics. They are dangerous when mathematics creates an illusion of certainty. An NPV of $438,721 is mathematically meaningful if the inputs are specified.

But it does not mean that the future value of the project is actually $438,721. It means that given a particular set of assumptions, the mathematical consequence is $438,721. That is a very different statement.

The distinction becomes increasingly important as forecasts stretch further into the future, terminal values become larger, discount rates become more consequential and human judgments become embedded in increasingly sophisticated models.

Forecasting error cannot be eliminated. Discount rates cannot be made purely objective. Terminal values cannot be known with certainty. Scenarios cannot capture every possible future. Behavioural biases cannot simply be removed by opening Excel. And more computation cannot compensate for assumptions that fundamentally misunderstand reality.

The answer is not to stop calculating. It is to become more intellectually honest about what calculation can and cannot accomplish.

A strong financial model should therefore do more than produce a number. It should expose uncertainty, challenge assumptions, identify vulnerabilities and help decision-makers understand how easily the answer can change.

The best model is not necessarily the one that gives the most precise answer. It is the one that makes it hardest to fool yourself. And that may be the most important lesson of all:

Mathematical precision is a property of the calculation. Predictive certainty is a property of the future. The first can be achieved. The second cannot.

Selected Research & References

U.S. Office of the Comptroller of the Currency — Model Risk Management: Revised Guidance — Current 2026 interagency guidance covering model development, validation, monitoring, governance, assumptions and model-use risk.
Aswath Damodaran — An Introduction to Valuation — Detailed discussion of valuation uncertainty, cash-flow forecasts and discount rates.
Aswath Damodaran — Terminal Value — Detailed treatment of terminal value and the sensitivity of valuation to perpetual-growth assumptions.
CFA Institute — Introduction to Financial Statement Modeling — Discussion of financial forecasting, long-term modelling and behavioural biases.
CFA Institute — Capital Market Expectations — Research-based discussion of forecasting limitations, data problems, model uncertainty and psychological biases.
NBER — Noise in Expectations: Evidence from Analyst Forecasts — Empirical evidence on forecast noise, bias and the deterioration of forecasting performance over longer horizons.
NBER — Man vs. Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases — Evidence concerning upward-biased earnings expectations and the relationship between forecast bias and horizon.
NBER — Belief Overreaction and Stock Market Puzzles — Research on excessive optimism in long-term earnings expectations and subsequent disappointment.
IMF — Model-Based Globally-Consistent Risk Assessment — Framework for representing uncertainty around forecasts and alternative scenarios.
IMF — Scenario Synthesis and Macroeconomic Risk — Recent work integrating scenario analysis with statistical forecasting and risk assessment.
Bank for International Settlements — To Count or Not to Count: The Future of Internal Models in Banking Regulation — Discussion of the limitations of financial risk models and the distinction between risk and uncertainty.
BIS — Prudent Valuation Guidance — Guidance emphasizing independent challenge of assumptions, model validation and recognition of model uncertainty.

When Numbers Lie: The Subtle Power of Financial Models and False Precision

Table of Contents

Quick Links

The Seduction of a Precise Number