AI | Treasury, FX, FI and MM
Written by Christopher Garner
Jan 25, 2021
Global financial markets are flooded with information, opinions and advice on the expected development of currency rates. Those responsible for managing FX exposures suffer from the headache that nobody can predict exchange rate movements dependably in the long-term. Consequently, FX hedging decisions frequently remain based on guesswork. This comes at the cost of forgone participation and unnecessary exposure to currency risk. What if there is a way to deal with the uncertainty of FX markets systematically and safely?
What do you do when there is a 40% chance of rain? Do you bring an umbrella; or do you choose the risk of being surprised by rain?
Humans must identify and manage risks every day in numerous situations. However, our brains face cognitive limitations affecting our behaviour, especially under uncertainty. These constraints often lead to overestimating low probability situations and underestimating real threats.
In such situations, there are opportunities for intelligent methods to assist in the decision-making process. Modern technology can deal with millions of data points and their probabilistic distributions far better and more consistent than humans can.
In the context of currency management, the key questions are how much to hedge and when?
LPA developed a system called “HedgePilot” to answer these challenges and support those tasked with managing foreign currency. Our solution considers each trade situation and systematically deals with FX market predictions to reduce unknown or unwanted currency risks. The AI-based algorithm continues its learning with new data, constantly improving decision-making quality.
HedgePilot is accessible via the web and can be used in addition to existing treasury tools. It provides tradeable signals based on the latest market data and consistent with any hedging policy and risk attitude a user might have.
Users can participate in beneficial market movements, while exposures are limited through a continuously monitored risk-limit enforced with a stop-loss. Essentially, we offer the payout profile of an FX-option without charging an expensive upfront premium.
AI is often referred to as a “Black Box“. However, due to recent research and developments in applied algorithms, this statement is no longer accurate. Given recent advances in AI, we can dissect the signals generated by HedgePilot and provide insights into the algorithm’s decision-making.
In contrast to a traditional economic model, with fixed variables and relations between them, our technology adapts to changing patterns and realities. Ultimately, traditional economic models are simplifications of reality, lacking robustness when faced with inadequate data. Subsequently, there has not been a single model that has successfully and consistently predicted FX movements. Rather than restricting ourselves through a pre-determined model with limited validity, we enable technology to search and identify patterns. This set-up is a systematic way of learning from the inherent uncertainties of market developments.
Based on LPA’s decades of expertise, we determine the playing field on which the algorithm is trained. We ensure that the input parameters, such as hedge strategy and hedge policy, always lead to signals that lie within plausible boundaries during the training process. Extreme market situations are simulated, assuring the algorithm learns from unobserved realities. As a consequence, the algorithm cannot veer onto unwanted paths, even during extreme situations.
Furthermore, rather than predicting an exact currency rate, we forecast probabilities (i.e. 40% of rain) and weigh various outcomes with possible client strategies. We embrace a systematic way of learning from the inherent uncertainties of market developments.
Safety features, comparable to driver-aids in modern automobiles, provide an additional layer of security:
Because of all of the steps undertaken in the algorithm’s development, and the fail-safes controlling the signals, we ensure that we do not receive implausible behaviour.
This quality applies to each signal generated through HedgePilot. Therefore, we believe that HedgePilot in the long-term will consistently beat any human-based decision – a development already seen in other industries, including chess and go.
Written by Theresa Jung
19 Sep, 2024
Written by Theresa Jung
18 Sep, 2024
Written by Susan Niederhöfer
1 Jul, 2024
Written by Günther Neurohr
26 Jun, 2024
Written by Patrick Reum
18 Jun, 2024
Written by Tatiana Karabatova
18 Jun, 2024