1 of 10

AI/ML APPLICATIONS IN CAPITAL MARKETS

Rachna Maheshwari

2 of 10

DISCUSSION AGENDA

2

  • Overview: ML in Capital Markets
  • Use Cases in Trading and Investment: Returns Forecasting & Trading Algorithms
  • Use Cases in Asset Management: Mutual Funds & Portfolio Optimization
  • Caveats and Risk Considerations

3 of 10

OVERVIEW: ML IN CAPITAL MARKETS

ML has applications in Portfolio Management and Trading

​

    • Forecasting Asset Returns
    • Selection of Funds & Fund Managers
    • Optimal Asset Allocation
    • Trading Strategies
    • Trading Algorithms
    • Trade Execution
    • Market Making
    • Sentiment Analysis

3

4 of 10

USE CASE #1: TRADING ALGORITHMS

4

  • Algorithms are widely applied in state-of-the-art trading by automating the trading process, using a set of if-but conditions. (Symbolic AI)
  • Use of Statistical AI, in trading involves embedding training of models within the algorithm and triggering trades based on model outcomes.
  • Implicit use of Statistical AI in trading is when ideas for trading strategies are obtained externally using AI, and conditions for triggering trades are defined based on the outcomes.
  • Explicit use of Statistical AI involves embedding ML within the trading algorithm.
  • Algorithms:
    • Classical Linear Regressions (Variations)
    • Decision Trees
    • Neural Networks (ANN, CNN, LSTM and MLP)
  • Input data:
    • Trading Data: Returns and Volatility
    • High Frequency Limit Order Book: Buy-Sell Orders, Bid-Ask Spreads, Trading Volumes, Price

(Source for high frequency LOB data: https://lobsterdata.com/)

    • Fundamentals: Accounting, Management & Sentiment Data
    • Macro: Macroeconomic & Financial Indicators
  • Conclusion:

Neural Networks and Deep Learning models outperform classical statistical and ML methods such as linear regression, and decision trees.

​

​

​

5 of 10

5

USE CASE #2 (1/2): FUND RETURNS FORECASTING

  • Regressions, Decision Trees and Neural Networks are used to forecast fund returns using data on fund specific and stock specific characteristics.

​

  • Various outcomes have been used for prediction:
    • Returns above the risk-free rate
    • Abnormal returns, which are returns not explained by risk factors (Alpha and Beta)

​

  • Data attributes that may be used in the algorithms:
    • Stock Holding Characteristics:
    • Returns & Profitability Measures: Returns Momentum, Return on Assets/Equity, Operating Profits, Margins, Fixed Costs to Sales Ratio.
    • Investments: Net Operating assets, Net Share Issues
    • Intangibles: Accruals, Operating Accruals, Operating Leverage, Price to Cost Margin
    • Value Measures: Asset to Market Capitalization Ratio, Book to Market Ratio, Cash Flow to Book Value etc.
    • Trading Frictions: Trading Volumes, Alpha, Beta, Volatility, Bid-Ask Spreads

​

    • Fund Based Variables:
    • Fund Characteristics: Fund Age, Total Net Assets, Fund Expense Ratio, Fund Turnover Ratio
    • Fund Momentum: Fund returns momentum – short term and long term
    • Fund Family Characteristics: Fund Family Age, Number of Funds in Family, Fund Family Total Net Assets

6 of 10

6

USE CASE #2 (2/2): FUND RETURNS FORECASTING

The rectified linear unit (ReLU) defined below is applied as the activation function on the input data:

​

The final output layer is simply a linear transformation of the output from the hidden layer, given by:

​

​

​

(Example taken from Kaniel R. et. al (2022))

7 of 10

USE CASE #3: PORTFOLIO WEIGHT OPTIMIZATION

7

https://quantpedia.com/hierarchical-risk-parity/

  • The objective function is to minimize the squared Euclidean distance between variable (Xs) data points across observations (i,j):

​

  • Data attributes used for this purpose are averages of returns, volatility, and measures of liquidity.

​

  • Unsupervised learning algorithm, such as hierarchical clustering analysis (Ward’s minimum variance method.).

​

8 of 10

CAVEATS & RISK CONSIDERATIONS

8

  • Risks associated with Model/Algorithm
    • Model explainability
    • Algorithm testing
    • Implementation and automation
    • Overfitting and model complexity

​

  • Risks associated with Data
    • Data quality and integrity
    • Data dimensionality
    • High frequency data and real time decision making

​

  • Governance & Regulatory Risks
    • Challenges with establishing a regulatory framework and detecting breaches
    • Algorithmic market manipulation
    • Concentration and procyclicality risks

​

​

9 of 10

9

REFERENCES

[1] Pinelis M., Ruppert D. (2022), “Machine Learning and Portfolio Allocation” The Journal of Finance and Data Science 8, 35–54.

[2] Kaniel R., Lin Z., Markus P., Van Nieuwerburgh S. (2022), “Machine Learning and the Skill of Mutual Fund Managers”, NBER Working Paper 29723.

[3] Briole A., Turiel J, Marcaccioli R., and Aste T., (2021), “Deep Reinforcement Learning for Active High Frequency Trading”, Working Paper, UCL.

[4] Karatas T., Malhotra S., (2020), “ML Applications in Asset Management” Presentation, ML in Finance Workshop, Columbia University.

10 of 10

THANK YOU!

PRESENTATION TITLE

10