Abstract:We address the problem of estimating the expected shortfall risk of a financial loss using a finite number of i.i.d. data. It is well known that the classical plug-in estimator suffers from poor statistical performance when faced with (heavy-tailed) distributions that are commonly used in financial contexts. Further, it lacks robustness, as the modification of even a single data point can cause a significant distortion. We propose a novel procedure for the estimation of the expected shortfall and prove that it recovers the best possible statistical properties (dictated by the central limit theorem) under minimal assumptions and for all finite numbers of data. Further, this estimator is adversarially robust: even if a (small) proportion of the data is maliciously modified, the procedure continuous to optimally estimate the true expected shortfall risk. We demonstrate that our estimator outperforms the classical plug-in estimator through a variety of numerical experiments across a range of standard loss distributions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to use a limited number of independent and identically distributed data to estimate the Expected Shortfall (ES) risk in financial risk management. Specifically, traditional interpolation estimators exhibit poor statistical performance in the face of heavy - tailed distributions (such as those commonly found in the financial field) and lack robustness. Even the modification of a single data point can lead to significant distortion. Therefore, this paper proposes a new estimation method, aiming to overcome these problems, achieve optimal statistical properties, and remain robust when the data is maliciously modified. ### Main contributions: 1. **Propose a new estimator**: The author proposes a new estimator $\hat{S}_N$, which restores the optimal statistical properties determined by the central limit theorem under minimal assumptions and is applicable to all finite amounts of data. 2. **Adversarial robustness**: The new estimator has adversarial robustness. Even if a small portion of the data is maliciously modified, the estimator can still optimally estimate the true expected loss risk. 3. **Numerical experiment verification**: Through a series of numerical experiments under standard loss distributions, it is proved that the new estimator is superior to traditional interpolation estimators. ### Specific problem description: - **Limitations of traditional interpolation estimators**: The traditional interpolation estimator $\hat{T}_N$ performs poorly in the face of heavy - tailed distributions and is very sensitive to the modification of a single data point. - **Advantages of the new estimator**: The new estimator $\hat{S}_N$ is not only superior to the traditional estimator in statistical performance but also can maintain good estimation performance when the data is maliciously modified. ### Mathematical background: - **Definition of expected loss**: For a random financial loss $X$, the expected loss $\text{ES}_\alpha(X)$ is defined as: \[ \text{ES}_\alpha(X):=\frac{1}{\alpha}\int_{1 - \alpha}^1\text{VaR}_u(X)\,du \] where $\text{VaR}_u(X):=\inf\{t\in\mathbb{R}:P(X\leq t)\geq u\}$ is the value - at - risk of loss $X$ at level $u$. - **Interpolation estimator**: The traditional interpolation estimator $\hat{T}_N$ approximates $\text{ES}_\alpha(X)$ through the empirical distribution function $\hat{F}_N(t):=\frac{1}{N}\sum_{i = 1}^N1_{(-\infty,t]}(X_i)$: \[ \hat{T}_N:=\frac{1}{\alpha}\int_{1 - \alpha}^1\text{VaR}_u(\hat{F}_N)\,du \] ### Construction of the new estimator: - **Block estimation**: Divide the data into several blocks, and use the interpolation estimator $\hat{T}_{I_j}$ for each block to estimate. - **Linear interpolation**: Perform linear interpolation on the estimation results of all blocks to obtain the empirical quantile function $\hat{Q}(\beta)$. - **Final estimation**: Combine the interpolation results and hyperparameters $\beta_1$ and $\beta_2$ to define the final estimator $\hat{S}_N$: \[ \hat{S}_N:=\min\left\{\max\left\{\hat{T}_N,\hat{Q}(\beta_1)\right\},\hat{Q}(\beta_2)\right\} \] ### Theoretical results: - **Statistical performance**: The new estimator $\hat{S}_N$ achieves optimal statistical performance under minimal assumptions, and the upper bound of the error is: \[ P\left(\left|\h

Optimal nonparametric estimation of the expected shortfall risk

Conditional Tail-Related Risk Estimation Using Composite Asymmetric Least Squares and Empirical Likelihood

Nonparametric Estimation of Expected Shortfall

A new non-parametric estimation of the expected shortfall for dependent financial losses

Nonparametric Estimation of Expected Shortfall for Α -Mixing Financial Losses

Nonparametric Expectile Shortfall Regression for Complex Functional Structure

Two-step online estimation and inference for expected shortfall regression with streaming data

Nonparametric Estimation of Expected Shortfall Via Bahadur-type Representation and Berry–Esséen Bounds

Robust Estimation and Inference for Expected Shortfall Regression with Many Regressors

Modified Marginal Expected Shortfall under Asymptotic Dependence

Extreme expectile estimation for short-tailed data

Marginal expected shortfall inference under multivariate regular variation

Nonparametric expected shortfall forecasting incorporating weighted quantiles

k-Nearest Neighbors Estimator for Functional Asymmetry Shortfall Regression

Robust Estimation of Operational Risk

Online Estimation and Optimization of Utility-Based Shortfall Risk

A novel scaling approach for unbiased adjustment of risk estimators

Recursive Estimation of the Expectile-Based Shortfall in Functional Ergodic Time Series

Tilburg University Estimation of the Marginal Expected Shortfall

From Data to Decisions: Distributionally Robust Optimization Is Optimal

Anytime-Valid Generalized Universal Inference on Risk Minimizers