Skip to main content
diversification.com
← Back to L Definitions

Likelihood

What Is Likelihood?

Likelihood, in the realm of statistical inference, is a measure of the plausibility of a specific statistical model or a set of its parameters, given a set of observed data. Unlike probability, which quantifies the chance of an outcome given a model, likelihood quantifies how well a particular model or parameter value explains the observed data. In quantitative finance, understanding likelihood is crucial for building robust financial modeling and interpreting analytical results. It helps in assessing which underlying conditions or parameters are most consistent with the market data observed. The concept of likelihood is fundamental to parameter estimation and model selection across various financial applications.

History and Origin

The modern concept of likelihood, particularly its application in maximum likelihood estimation (MLE), was largely developed by the British statistician and geneticist Sir Ronald Fisher in the early 20th century. Fisher's seminal work between 1912 and 1922 laid the foundation for using likelihood as a central tool in statistical theory. While he introduced the method in 1912, the term "likelihood" itself was coined by him later. His work profoundly influenced how scientists and researchers approach data analysis, moving towards more rigorous methods for inferring underlying truths from observed phenomena.4

Key Takeaways

  • Likelihood measures the plausibility of a model or parameter value given observed data.
  • It is distinct from probability, which quantifies the chance of an outcome.
  • Maximum Likelihood Estimation (MLE) is a widely used method for estimating model parameters.
  • Likelihood plays a critical role in various financial analyses, from risk assessment to portfolio construction.
  • Interpreting likelihood values requires understanding the context of the underlying statistical model.

Formula and Calculation

The likelihood function, $L(\theta|x)$, expresses the likelihood of the parameters $\theta$ given the observed data $x$. For independent and identically distributed (i.i.d.) observations, the likelihood function is the product of the probability density function (PDF) or probability mass function (PMF) for each observation.

For a set of observed data points $x = (x_1, x_2, \ldots, x_n)$, and a statistical model with parameters $\theta$, the likelihood function is given by:

L(θ∣x)=∏i=1nf(xi∣θ)L(\theta|x) = \prod_{i=1}^{n} f(x_i|\theta)

Where:

  • $L(\theta|x)$ is the likelihood of the parameters $\theta$ given the data $x$.
  • $\prod_{i=1}^{n}$ denotes the product over all $n$ observations.
  • $f(x_i|\theta)$ is the probability distribution (PDF or PMF) of a single observation $x_i$ given the parameters $\theta$.

Often, it is more convenient to work with the log-likelihood function, which converts the product into a sum, simplifying calculations, especially in maximum likelihood estimation:

ln⁡L(θ∣x)=∑i=1nln⁡f(xi∣θ)\ln L(\theta|x) = \sum_{i=1}^{n} \ln f(x_i|\theta)

Interpreting the Likelihood

Interpreting likelihood involves assessing which parameter values or models best explain the observed data. A higher likelihood value for a given parameter set suggests that those parameters are more consistent with the data collected. It is not an absolute probability of the parameters being true, but rather a relative measure of support provided by the data for different parameter values within a model. In quantitative finance, this often means identifying the most plausible values for factors like expected returns, volatility, or correlation, given historical market movements. For example, when estimating parameters for an asset pricing model, comparing the likelihood of different parameter sets helps determine which set offers the best fit to observed asset prices. This interpretation is key for informed decision-making in areas like risk management.

Hypothetical Example

Consider an investor trying to determine the average daily return ($\mu$) of a particular stock, assuming its daily returns follow a normal distribution with a known standard deviation ($\sigma$). Over 10 days, the observed daily returns are: +0.5%, -0.2%, +1.0%, +0.1%, -0.5%, +0.8%, +0.3%, -0.1%, +0.6%, +0.2%.

To find the likelihood of different possible values for $\mu$, the investor would calculate the likelihood function for each hypothesized $\mu$, given the observed returns and the known $\sigma$. For instance, if the investor assumes $\sigma = 0.5%$ and wants to compare $\mu = 0.2%$ versus $\mu = 0.4%$:

The likelihood for each observation $x_i$ for a normal distribution is given by:

f(xi∣μ,σ)=12πσ2exp⁡(−(xi−μ)22σ2)f(x_i|\mu, \sigma) = \frac{1}{\sqrt{2\pi\sigma^2}} \exp\left(-\frac{(x_i - \mu)^2}{2\sigma^2}\right)

The total likelihood is the product of these individual likelihoods. The investor would calculate this product for $\mu = 0.2%$ and then for $\mu = 0.4%$. The mean of the observed data (0.27%) would likely yield the highest likelihood value, as maximum likelihood estimation often converges to the sample mean for normally distributed data. This process helps to select the parameter value that best explains the observed return data.

Practical Applications

Likelihood is a cornerstone in numerous practical financial applications, especially in quantitative finance and risk management. It is extensively used in:

  • Model Estimation: From calibrating option pricing models to estimating parameters for econometric models that describe asset behavior, likelihood functions are used to find the parameters that best fit historical data.
  • Credit Risk Modeling: Financial institutions use likelihood to estimate the probability of default for loans or bonds, fitting models to historical default data to determine the most likely parameters for future credit events.
  • Recession Forecasting: Central banks and economists employ models that use likelihood to estimate the economic forecasts of a recession occurring within a certain timeframe. The Federal Reserve Bank of San Francisco, for example, publishes analyses on the likelihood of recession based on various economic indicators.3
  • Portfolio Optimization: While classic mean-variance portfolio optimization uses expected returns and covariance, more advanced techniques might use likelihood to estimate underlying return distributions and their parameters, leading to more robust portfolio allocations.
  • Financial Distress Prediction: Models predicting corporate bankruptcy or financial distress often rely on likelihood methods to estimate the parameters that best distinguish between healthy and distressed firms based on financial ratios.
  • Policy Evaluation: Regulatory bodies may use likelihood-based approaches to evaluate the effectiveness of new policies or regulations by assessing how well a model with the new policy parameters explains observed market outcomes. For example, during discussions about the U.S. debt ceiling, various scenarios are analyzed to determine the likelihood of a government default, influencing policy decisions to avert such an outcome.2

Limitations and Criticisms

Despite its wide applicability, likelihood-based methods have limitations. A significant criticism revolves around the assumption of a correctly specified model. If the underlying statistical model chosen does not accurately reflect the true data-generating process, the likelihood estimates, particularly in maximum likelihood estimation, may be biased or inefficient. Furthermore, likelihood functions can be computationally intensive for complex models or large datasets, requiring sophisticated optimization algorithms.

Another challenge arises in highly volatile or uncertain markets. Models relying on historical data to derive likelihoods might fail to adequately capture extreme or unprecedented events, leading to inaccurate risk management assessments. For instance, in portfolio optimization, relying solely on historical likelihoods for expected returns can make portfolios vulnerable to unexpected market shifts. This has led to the development of alternative approaches, such as robust optimization, which explicitly account for uncertainty and ambiguity in model inputs rather than assuming a fixed underlying distribution.1 These methods aim to create portfolios that are less sensitive to estimation errors and more resilient across a wider range of possible scenarios.

Likelihood vs. Probability

While often confused, likelihood and probability are distinct concepts in statistics and quantitative finance. Probability refers to the chance of a specific outcome or event occurring, given a known model or set of parameters. For example, the probability of flipping a fair coin and getting heads is 0.5.

In contrast, likelihood refers to how plausible a given set of parameters is, given the observed data. It answers the question: "Given this data, how well do these parameters explain it?" For instance, if you observe 7 heads out of 10 coin flips, the likelihood of the coin being fair (probability of heads = 0.5) might be calculated, as well as the likelihood of it being biased (e.g., probability of heads = 0.7). The likelihood function would show which parameter value (0.5 vs. 0.7) provides a better explanation for the observed 7 heads. While probability integrates over outcomes, likelihood integrates over possible parameter values.

FAQs

What is the primary difference between likelihood and probability?

Probability quantifies the chance of observing specific data given a known model or its parameters. Likelihood, on the other hand, quantifies how well a particular model or set of parameters explains the observed data. It's about evaluating the parameters given the data, rather than evaluating data given the parameters.

How is likelihood used in financial markets?

In financial markets, likelihood is used to estimate unknown parameters in financial models, such as the volatility of asset prices, the parameters of asset return distributions, or the relationships between different economic variables. This is crucial for applications like risk management, option pricing, and portfolio construction.

Can a likelihood value be greater than one?

Yes, unlike probability, a likelihood value can be greater than one. This is because the likelihood function is not a probability density function for the parameters; it does not need to integrate to one. Its purpose is to show the relative plausibility of different parameter values given the data.

What is Maximum Likelihood Estimation (MLE)?

Maximum Likelihood Estimation is a method of estimating the parameters of a statistical model. It finds the parameter values that maximize the likelihood function, meaning they are the parameter values under which the observed data is most probable. These estimates are often preferred due to their desirable statistical properties, such as efficiency and asymptotic normality, especially when sample sizes are large.

How does likelihood relate to confidence intervals?

Likelihood is foundational for constructing confidence intervals. Confidence intervals provide a range of values within which a true population parameter is likely to lie. These intervals are often derived from the properties of maximum likelihood estimators, leveraging the shape of the likelihood function to determine the uncertainty around a parameter estimate.