SAMLAF: Security Assessment of Machine Learning Applications in Finance

back to overview

Type and Duration

FFF-Förderprojekt, September 2023 until February 2025


Data & Application Security

Main Research

Business Process Management


Algorithmic trading has become a vital instrument in the financial services industry. Automatic decision making on financial markets with the help of intelligent algorithms enables traders to increase the volume of their operations and hence to optimize their profits. Machine Learning (ML) is becoming increasingly popular in algorithmic trading as it enables decisions to be driven by large volumes of data rather than hardcoded rules. Advances in computational intelligence revealed that ML methods can yield a superior predictive performance in the context of algorithmic trading in comparison to traditional time-series forecasting (TSF) methods (e.g., ARIMA). Despite such advantage, deployment of ML exposes algorithmic trading systems to the risk of adversarial examples, i.e., small perturbations to input data that cause substantial prediction errors. In this project, we scrutinize the security of ML applications for TSF in finance and, specifically, for algorithmic trading. The necessity of this research arises from the peculiarity of algorithmic trading which attempts to identify microtrends in the behavior of financial time series which are based on publicly available data, and hence exploitable by everybody. Prior work on security of ML has hardly considered such deployment scenarios, and the few recorded accounts did so by making far-fetched assumptions. Hence the proposed research is aimed at: (i) uncovering the real risks related to deployment of ML for financial predictions, (ii) potentially quantifying such risks, and (iii) assessing feasible countermeasures if required.

Practical Application

All the experiments carried out in this project will be based on real data, which will be used to develop ML models that exhibit a predictive performance that is useful in practice. As such, it is reasonable to assume that all the findings of this research will allow to estimate the real threat of adversarial attacks against ML systems for financial predictions. Therefore, our findings will have a high value to practitioners within the financial domain.

Reference to Liechtenstein

Liechtenstein will greatly benefit from the SAMLAF project, since it deals with themes (artificial intelligence, finance, cyber security, and information systems) with abundant interest to Liechtenstein's development, and which also represent a core component in the University of Liechtenstein's portfolio. In particular, the biggest opportunity for Liechtenstein lies in the the novelty of the SAMLAF project: its accomplishment would mean that Liechtestein is the first country to carry out sound research on the security of artificial intelligence applications in finance.


Cybersecurity Management, Finance

Principal Investigator

Project Collaborator