Which ML model is best for stock prediction?

March 16, 2023

Which ML model is best for stock prediction?

Artificial Intelligence & Machine Learning

How do you convince a discretionary Portfolio Manager to use Machine Learning as part of their investment decision making process?

My thinking process behind creating the presentation deck.

1. I will not convince the Portfolio Manager that Quantamental Investing is better than Discretionary Investing, rather I will first understand their perception about Quantamental Investing and then develop the content accordingly

2. Moreover, the approach that I will take to shift their perspective is to first get into the shoes of the Portfolio Managers to understand their viewpoint on Quantamental Investing. I will then bridge the gap so that they can reach the place that I am at.

3. I firstly understood the “Why” behind why the Portfolio Managers aren’t using Quantamental Investing – a. They don’t know about it!

b. They know about it but haven’t got the chance to explore it.

c. They want to explore it but don’t know how to adopt it.

d. In addition, they don’t trust the quantamental process.

4. I then curated the content of the slides keeping in mind what it would take to influence the decision of these 4 categories of Portfolio Managers and tailored the content to cater to their needs.

5. I will finally describe a unique investment idea using machine learning

Fundamentals of Quantamental Investing in addition, the use of machine learning & mathematical modeling!

What is it?

Present State & Practical Uses

Furthermore, modeling, and data analysis to calculate the optimal probability of executing a profitable trade.

Quantamental funds run around 35% of the $31 trillion of equity assets.

Human managers, such as traditional hedge funds and other mutual funds, manage just 24%.

Using machine learning, advanced mathematical modeling, factor investing and the use of alternative data, to make trading and portfolio allocation decisions.

Performance of Quantamental Investment

Broad market factors such as value, growth, quality, and momentum have contributed to 65% of global equity manager’s excess returns.

“Three years ago, quant funds became the largest source of institutional trading volume in the American stock market. They account for 36% of institutional volume so far this year, up from just 18% in 2010, according to the Tabb Group.”

The Quantamental Investing Process

Data Sources

Gather different types of data streams to identify hidden signals.

Convert them into factors with economical meaning.

Model & Factors

Combine the factors into composite signals through analytic tools. In addition, analyze the efficacy of factors explaining the movement of stock prices to check if they have persistent value over time

Alpha Generation

Forecast the portfolio measures using the models. Compute the forward returns for out of sample data

Analyzed these portfolios alongside more traditional factor-based portfolios.

Portfolio Execution

Transform predictions and signals into portfolios decisions using quantitative methods.

Build optimized portfolio from these various models giving different signals.

Post Trade Analysis

Measure the performance of the portfolio after simulating trades, using established metrics to provide feedback to model building.

Benefits of Quantamental Investing : Improved accuracy and efficiency of trade executions.

Can consider liquidity, integrate risk-assessments and impact of large trades on the market.

Exploiting alternate forms of data to produce better alpha.

ML tools can compile geolocation data, satellite imagery, and macroeconomic data.

Eliminates human-errors such confirmation bias, emotions, loss-aversion.

Furthermore, ML can be used for security selection, portfolio construction and trading executions.

ML creates adaptive portfolios.

ML inputs decide how to weight allocations within an index to favor certain characteristics like low volatility.

Limitations of Quantamental Investing

Black Box: Skilled personnel & Data

Scientists are required to develop and explain complex models to investors. Lack of understanding behind the signals, judgements and strategy makes the investors uncomfortable_.

Investors do not adapt well to higher sigma moves or black swan events, if they have no historical data of such events.

Overfitting: Wrong model selection and training methods can give poor results in the market. Not eliminating the noise can give data-mining bias

Data Sources: Obtaining quality, relevant and reliable data is crucial in training & testing the model. Using old data leads to incorrect model predictions.

Need for Alternate Investing Techniques

The Present state

Only a handful of financial institutions have incorporated a quantitative approach to investing.

Future Scope

AI and Big Data, when used in conjunction, will improve the efficiency of quantitative models by accurately accounting for alternate data

The Problem

Bottom-up analysis overlooks macro-trends and is influenced by human emotions. Traditional

methods overlook new sources of data.

The Solution – Develop a quantitative approach to investing

Identify new alternate data sources (like credit cards, receipts, GPS data, textual data) to produce robust ML models

The Unique Idea – Sentimental Analysis of Management Discussions & Analysis
The Intuition – Analyze the company’s performance through the eyes of management to make better investment decisions
The Approach – Ordinal classification of sentiments behind non-financial data through automatic document classification
The Models – The model consists of three integrated submodules (using NLP) to identify historical trends and patterns to enhance investment making

1. Data Acquisition Model, which will gather MD&A data & clean it to produce machine comprehensive inputs 2. Score-Based Intent Analyzing Model, which will identify inputs and assign a score, based on a predefined intent gauging spectrum. Can incorporate contextual semantic search, to smartly segregate meaning behind the message without programming

3. Trade Analysis Model, which will group the data into major baskets for decision making. Baskets which cross a score threshold will raise flags, requiring attention for human intelligence and nuance.

Investment Strategy – We incorporate the quantitative signals of the models to make better estimates of a company’s future. We identify undervalued equity stocks through our ML Models and seek higher return