This talk will discuss how Bloomberg machine learning projects - such as sentiment analysis, market impact prediction, recommender systems, social media analytics and liquidity analysis - are practically applied to key problems in the financial industry: strategy development, portfolio optimization, knowledge discovery, and others. We will discuss the recent evolution of the machine learning landscape from the perspective of the participants in the global financial industry, summarize the current state of the art and discuss plausible future directions for the applications of machine learning methods in finance.
Machine Learning Applications at Bloomberg
Gary Kazantsev, Bloomberg
Multi-asset class longer-horizon investors face different modeling challenges from shorter-horizon traders with regard to liquidity, regime shifts, and drift. We describe an empirical framework for estimating portfolio return distributions which enable a more systematic approach to constructing multi-asset class portfolios that are more robust to macroeconomic shocks and changing volatility regimes.
A Data-Driven Approach to Multi-asset Class Portfolio Risk Estimation
Shota Ishi, State Street GX
With a daily volume of thirty to fifty million US dollars and a market capitalization over five billion, Bitcoin is becoming interesting as a financial instrument for inclusion in a quantitative trading strategy. We will explore the unique issues of the various exchanges, impact of exogenous events and demonstrate a fully automated machine learning based trading system.
Machine Learning Based Bitcoin Trading
Arshak Navruzyan, Startup.ML
The speaker estimates that the majority of quantitative strategies which are traded are based on biased and broken backtests and questionable research, and many if not most have negative expected returns. In this talk, some sources of these errors, and potential cures are discussed. This rogue's gallery of broken backtests were accumulated over 8 years by the speaker during his tenure as a quantitative strategist at two machine-learning based hedge funds.
Guarding Against Broken Backtests and Questionable Research in Quantitative Strategies
The speaker will introduce the open source python libraries Zipline and Pyfolio which can be used for backtesting and analyzing algorithmic trading strategies. The talk will focus on stepping through an example zipline implementation of a trading strategy based on machine learning. Discussion will evolve into how to use the example as a template for further exploration and potential paths for building out increasingly more sophisticated trading algorithms. Finally, the performance of a trading algorithm will be analyzed using pyfolio, including explanations of its many visualizations, and how the data presented can be used to evaluate whether a trading algorithm may be viable for actual investment.
Deep neural networks (DNNs) are powerful types of artificial neural networks (ANNs) that use several hidden layers. They have recently gained considerable attention in the speech transcription and image recognition community (Krizhevsky et al., 2012) for their superior predictive properties including robustness to overfitting. However their application to algorithmic trading has not been previously researched, partly because of their computational complexity. This talk discusses our feedback from working with algo-trading firms to put these methods into a live simulation environment. In particular we cover quality and choice of data sources, forecasting horizon, frequency of retraining, reinforcement methods, trade expression, operational risk management - all important considerations which impact the profitably of a ML based trading strategy.
Classification-based Financial Markets Prediction using Deep Neural Networks
Matthew Dixon, Illinois Institute of Technology