The conference is going to take place at Aula Magna, Faculty of Law, Egyetem tér 1-3., district 5, Budapest, Hungary, H-1053.
Existing hedging strategies are typically based on specific financial models: either the strategies are directly based on a given option pricing model or stock price and volatility models are used indirectly by generating synthetic data on which an agent is trained with reinforcement learning. In this research, we train an agent in a pure data-driven manner. Particularly, we do not need any specifications on volatility or jump dynamics but use large empirical intra-day data from actual stock and option markets. The agent is trained for the hedging of derivative securities using deep reinforcement learning (DRL) with continuous actions. The training data consists of intra-day option price observations on S&P500 index over 6 years, and top of that, we use other data periods for validation and testing. We have two important empirical results. First, a DRL agent trained using synthetic data generated from a calibrated stochastic volatility model outperforms the classic Black-Scholes delta hedging strategy. Second, and more importantly, we find that a DRL agent, which is empirically trained using actual intra-day stock and option prices directly without the prior specification of the underlying volatility or jump processes, has superior performance compared with the use of synthetic data. This implies that DRL can capture the dynamics of S&P500 from the actual intra-day data and to self-learn how to hedge actual options efficiently.
There is a long standing belief that a discipline becomes a science when it can be mathematized. With the advent of AI, the concept of science is widely expanded, with social science being a likely winner. This talk will present a possible view of the future of a largely expanded financial sector drawn by technology and how social science can become a new partner.
In the talk a novel method is presented to extract relevant feature information for time series. We find patterns in the form of (linear) laws, and characterize the time series with the best fitting laws. The resulting Linear Law based feature Transformation (LLT) makes classification tasks more effective, as it will be demonstrated in some examples, including financial data analysis.
q/kdb+stands as the preeminent time series analysis tool in the global capital market, boasting unrivaled speed and efficiency over the past three decades. Its distinguishing features, including vector and functional programming, alongside native data tables with an extended
SQL, have cemented it as a foundational language for quantitative analysts. In 2022, KX introduced
PyKX, a seamless integration tool that empowers
Pythondevelopers to harness the power of
qproficiency. Furthermore, the introduction of
KDB.AIrevolutionizes knowledge-based vector databases, enabling developers to construct scalable, reliable and real-time applications by providing advanced search, recommendation and personalization for AI applications.
Identifying companies with similar profiles is a core task in finance with a wide range of applications in portfolio construction, asset pricing and risk attribution. When a rigorous definition of similarity is lacking, financial analysts usually resort to ‘traditional’ industry classifications such as Global Industry Classification System (GICS) which assign a unique category to each company at different levels of granularity. Due to their discrete nature, though, GICS classifications do not allow for ranking companies in terms of similarity. In this paper, we explore the ability of pre-trained and finetuned large language models (LLMs) to learn company embeddings based on the business descriptions reported in SEC filings. We show that we can reproduce GICS classifications using the embeddings as features. We also benchmark these embeddings on various machine learning and financial metrics and conclude that the companies that are similar according to the embeddings are also similar in terms of financial performance metrics including return correlation.
Introduction to Morgan Stanley Wealth Management journey to build an Intelligent organization. In the past years we focused on scaling up our Machine Learning capabilities, building several tools to help the business. I will walk through the audience on 3 practical actual use-Case where Machine Learning was implemented to data products solving the problem on digital client engagement, Advisors engagement towards clients and on how to match advisors better with clients. We will cover the business problem, the concept of the solution and finally the key learning for each use-cases.
We train a scale-free LSTM-type neural network on a massive amount of fractional Brownian motion or fractional Ornstein Uhlenbeck process trajectories to learn the Hurst exponent of those processes. While the network’s performance is excellent in terms of the mean squared error, the absolute and relative error quantiles are substantial due to a skewed distribution. True, though, the network still overperforms the traditional statistical Hurst estimators. There is a line in the literature advocating for a fractional-Brownian-motion-based modeling of the S&P500 index. By conditionally accepting that model, we illustrate the effect of the network’s misspecification of the Hurst exponent on option pricing. We present the actual calculations on two-days-to-maturity call prices of Nov 3, 2023.
Portfolio managers often need to solve optimization problems to determine the ideal allocation of their managed accounts. Depending on their strategy of choice this can involve dealing with difficult mathematical models. Traditionally these are handled by solving a sequence of easier subproblems, where each subproblem is determined by a heuristic step based on information already obtained during the process. Recently there is more and more interest in applying machine learning methods to deal with these problems. In the talk, we will give an overview of certain optimization problems and discuss some ways machine learning can play a part in the process of solving them.
We map one-dimensional quantum systems to classical time series and explore how strong quantum correlations turn into nontrivial classical time autocorrelations. In particular we show that the Luttinger liquid properties of quantum magnets translate into multifractal time series characterized by non-trivial Hurst exponents. We show that the classical series can be sampled sequentially from the known or numerically determined Matrix Product State approximation of the quantum ground state.
Citi has developed a machine learning model to predict vol surface deformation scenarios on quarterly earnings of US single stocks derivatives market giving fast market color to traders.
The workshop is free for registered participants. You can register until Nov 19, 2023. If you want to cancel your registration contact the organizers at firstname.lastname@example.org.
With my registration I give consent to and permit image- and sound-recording to be taken of me on the event, and these recordings to be used by the organizer(s) in their internal and external communications (e.g. with aims as reporting and giving information about the event, propagating/publicizing the event, using them as reference).
These recordings of me can be used for the above mentioned goals by any media provider free of charge, without any place or time limitation, through any technology suitable for broadcasting to the public, without any limitations regarding the number of times being used, and through every known utilization method stated in the Act LXXVI of 1999 on Copyright.
With my registration, I give permit to store and use my data during the organization of the current and future workshops. These data will not be shared with any third parties.