kinetic mind
quantitative finance meets predictive analytics
kinetic mind specializes in developing and implementing intelligent processes for trading, risk control and portfolio management by combining quantitative finance with predictive analytics and machine learning concepts
Predictive Analytics
Statistical models based on neural networks allow to make sound predictions which help making better business decisions today.
Typical examples are:
- Predicting 1/4-hourly electricity prices for optimizing power plant load schedules using market and fundamental data.
- Predicting future cash flows to improve the FX liability management.
Engineering
Combining classical numerical methods with statistical learning techniques pioneer new and innovative solutions to common problems in engineering.
Use cases are:
- A machine learning based recommender system that prepares promising proposals for arbitrary parameters in partial differential equation (PDE) problems.
- An autonomous control unit based on reinforcement learning techniques that continuously learns from feedback and is able to handle non-standard situations.
Automated Hedging
Modern concepts that employ machine learning techniques advance trading and risk management operations.
Such techniques may be used:
- To compute hedge quantities for a portfolio when transaction costs and liquidity constraints are present.
- To set up portfolio management strategies that benefit from taking a specific risk but need to hedge all remaining risks.
Physical Assets
Operating physical assets like power plants or gas storages can be very challenging in view of maximizing cash flows, while minimizing risks and simultaneously allowing for technical restrictions.
Intelligent systems may support:
- Finding optimal proxy hedges for physical assets with no liquidly traded hedge instruments.
- Optimizing hedge positions and actions in the spot market with respect to trading costs and liquidity constraints.
- Identifying the fair risk premium with respect to individually defined risk measures.
Data Error Detection
The risk management of portfolios requires pre-processed market data (e.g. discount curves, forward curves and volatility surfaces). These bootstrapped inputs are often subject to errors arising from incorrectly finished processes, possibly causing severe losses.
Neural network models like Autoencoder provide a valuable tool for predictive error detection. By construction they validate the consistency of the entire data object going far beyond the usual plausibility checks.
Risk Monitoring
The responsibilities of modern risk control covers a wide area of operations. In particular you will find recurring and standardized tasks among them. These may be outsourced to intelligent autonomous agents.
Possible tasks that can be automated:
- The risk assessment of trade positions by using the hedge position of an automated hedger as the benchmark.
- The risk limit determination by using „greeks“ (delta, gamma, vega etc.) from an automated hedger.
- The evaluation of unhedgeable risks, e.g. correlation risk or risk arising from hourly structures in load profiles.