Projects | Sebastian Stöckl

Risk-adjusted Forecast Performance

Wed, 11 Feb 2026 00:00:00 +0000

When Forecast Accuracy Fails to Translate into Performance

Financial forecasting models are typically evaluated using statistical metrics such as RMSE or out-of-sample R². However, investors ultimately care about economic performance — e.g., Sharpe ratios, portfolio efficiency, and risk-adjusted returns.

This project studies the disconnect between statistical forecast accuracy and economic value.

Core Contribution

In (Journal of Forecasting, forthcoming), we:

Derive the Sharpe-ratio gap between perfect foresight and estimated portfolios
Decompose it into:
- Risk-Adjusted Mean Forecast Error (RAFE)
- Risk-Adjusted Covariance Forecast Error (C-RAFE)
Show that the widely used RMSE is a special case of a risk-adjusted loss function under highly restrictive assumptions
Demonstrate that RMSE explains little of the performance shortfall
Show that risk-adjusted error measures explain the Sharpe-ratio gap across a wide range of portfolio strategies

Key Insight

Forecast models optimized using MSE-based criteria are not aligned with the investor’s objective function.

Improving statistical fit does not necessarily improve economic performance.

Ongoing Research

This project extends toward:

Risk-adjusted loss functions in regression and machine learning
Portfolio-aware model selection
Utility-consistent forecast evaluation
AI-based asset allocation with economically aligned objectives

Decision Methods and Tools in the Context of Pension Finance

Fri, 04 Dec 2020 00:00:00 +0000

Overview

In this project we developed an R-package (available on GitHub at ) to optimize decisions individuals in Liechtenstein’s pension system have to take. The package contains several optimizers as well as full documentation (available via vignette("model") once installed). We determined the most relevant drivers of optimal pension decisions using large-scale optimization, then trained three machine learning models — a hyperparameter-tuned random forest performs best — to allow individuals to receive fast, near-optimal decisions. Results are publicly available at .

Step-by-Step Project Description

Built the R-package pensionfinanceLi (devtools::install_github("sstoeckl/pensionfinanceLi")), carefully validated to avoid counter-intuitive results.
Provided detailed documentation of the package and Liechtenstein pension system (also available as vignette("model")).
Determined plausibility checks for all input parameters, drew preliminary conclusions on parameter dependencies, and reduced undesired outcomes (e.g. moral hazard).
Compiled a final grid of 3,110,400 feasible parameter combinations for large-scale optimization.
Due to the time-consuming full-parameter optimization (~25 min/run), optimized 3 key decision variables: consumption share c, retained wealth share alpha, and optimal asset allocation w. Parallelized across Amazon AWS clusters (3 × 96 CPU cores) and continued on an in-house 40-CPU cluster.
Ran one- and two-dimensional comparative static analyses to identify the most relevant drivers of optimal pension decisions.
Trained three machine learning models in Python (linear regression, k-nearest neighbors, tuned random forest). The random forest achieves surprisingly good out-of-sample performance.
Deployed a public where anyone can input their individual situation and retrieve near-optimal pension decisions from the trained heuristics.

Graphical description of the project workflow.

Macro-Financial Uncertainty

Thu, 12 Dec 2019 00:00:00 +0000

Political Event Portfolios

Thu, 12 Dec 2019 00:00:00 +0000

Understanding Saving in Europe

Sun, 01 Sep 2019 00:00:00 +0000

Apply for free online courses at
Official link to

Understanding Pensions in Europe

Fri, 01 Sep 2017 00:00:00 +0000

Apply for free online courses at
Official link to