How Machine Learning Backed Rule-Based Investing  Is  Disrupting Active Fund  Management

FidelFolio smallcase Philosophy

Fidelfolio’s core philosophy emerged from the vision to revolutionize investment management industry by addressing its most significant obstacle head-on. Our experience in fund management, coupled with conversations with key industry players – fund managers, research analysts, and wealth managers, had given us comprehensive understanding of the pressing issues in investment research. These issues distill into 3 primary pitfalls – Human Biases, Limited Analytical Power, and Human Errors.

Among these challenges, human biases are probably the most detrimental. Despite our belief in rational decision making, psychological, emotional, and behavioral biases often influence our choices. Conquering these biases is a pivotal step toward enhancing investment research. Nevertheless, the other 2 challenges – Limited Analytical Power and Human Errors – trail closely behind in their potential to dent your investment gains and overall wealth. Limited analytical capabilities limit the breadth of investment opportunities, while human errors can often prove to be near-fatal blunders.

Exhibit 1: Challenges in investment research

To overcome these challenges, we have developed an innovative investment framework that combines the power of machine learning (ML) with human intelligence. By leveraging ML algorithms in conjunction with human expertise, we aim to eliminate biases, enhance analytical capabilities, and reduce the risk of errors in the investment research process.

Our approach involves the continuous refinement of investment strategies through the self-learning capabilities of our ML-powered strategy generator and analyser, while also ensuring human interpretability & validation, and transparency.

WHY NOW? – UNPRECEDENTED DATA CAPABILITIES

The emergence of artificial intelligence (AI) and machine learning (ML) is not a recent phenomenon. The term “artificial intelligence” was coined by John McCarthy in 1956, and ML has been a topic of discussion for decades. However, what has truly accelerated the adoption of ML in recent years is the remarkable increase in data processing

capabilities. Today, we can process data thousands of times faster than we could just a decade ago. This exponential growth in data processing power has paved the way for the widespread application of ML and AI across various industries, including investment management.

MOST POWERFUL TOOL EVER?

The impact of ML and AI in investment research and management processes has far surpassed the effects of previous tools such as paper and pencil, calculators, and computers combined. While earlier tools were limited to record-keeping and basic calculations, AI and ML can now conduct complex analytics, self-learning, idea generation,

and in-depth pattern recognition that surpass human capabilities. This has resulted in a significant shift in the role of humans in investment management, where their bandwidth is freed up for qualitative analysis and decision- making, while the ML algorithms handle the data-intensive tasks.

 Exhibit 2: Impact of ML in Investment Research Process

PENETRATION OF ML IN INVESTMENT INDUSTRY

We can observe a clear trend towards the integration of ML into various investment approaches – including short- term trading, long-term investment, technical and fundamental analysis. Traditional human-only approaches, particularly those focused on short-term trading using technical analysis, are becoming obsolete. High-frequency trading (HFT), quant hedge funds, and trading algorithms driven by ML are replacing these outdated methods.

Meanwhile, a new wave of investment firms has emerged, leveraging ML in combination with fundamental analysis to create investment strategies that are not solely focused on short-term gains. Most of these firms struggle to strike a balance between not being a black box algorithm and maintaining a low- churn investment product that can operate at scale.

Exhibit 3: Penetration of ML in Investment Management

OUR INVESTMENT FRAMEWORK

FidelFolio recognizes the importance of addressing these challenges and has developed an investment framework that combines machine learning (ML) with human intelligence. The framework comprises two pillars: the strategy generator and the strategy analyser, which create and back-test investment strategies in a self-learning mode.

The human component plays a vital role in training the algorithms, validating recommendations, conducting qualitative analysis, and ensuring the understandability & practicality of the recommendations. FidelFolio’s approach is not a black box algorithm but one that can be comprehended and monitored by humans.

Exhibit 4: Fidelfolio’s Investment Framework

Let’s delve into a practical example of our ML engines at work, where a single investment rule is conceived and subjected to backtesting. The process commences with our investment team imparting training to the FF Strategy flenerator, facilitating the creation of transparent investment rules grounded in fundamental ratios. To illustrate, consider the inception of the initial rule—’13-14- 24-20-8-11-C’. This signifies the selection of companies that have achieved a minimum of 20% Return on Equity and 8% growth in Operating Profit for at least 11 out of the last 13 years. The FF Strategy flenerator diligently sifts through the stock list each year over the past 32 years, sieving out those that align with this investment criterion. Thus, we have the historical roster of portfolio constituents and modifications in accordance with this rule spanning the past three decades.

But what do we do with this investment-rule and its historical details? Obviously, our system needs to know if the rule created is good enough for investment or not. This is where the FF Strategy Analyser comes into play. Our investment team has diligently trained this analyser to rigorously backtest the investment rule – a task made feasible by our comprehensive historical portfolio records linked with this specific rule. The outcome of these backtests reveals that investing according to this rule has yielded an average annual return of 20%, a standard deviation (risk) of 25%, sortino ratio of 1.4x, and while the most exceptional one-year return reached an impressive 120%, with the lowest plummeting to -30%. Assuming we have defined Nifty50 as our benchmark, our system evaluates each metric by comparing it with Nifty50 and categorise it as ‘favourable’ or ‘unfavourable’.

OUR INVESTMENT PROCESS

To understand FidelFolio’s Investment process, let us continue with the previous example of formulating, backtesting, and assessing a single investment rule using our system. This evaluation serves as a feedback mechanism, essentially ‘informing’ the FF Strategy glenerator of the quality of its performance.

Based on this feedback, the generator adapts itself to produce improved investment rules. The subsequent analysis and evaluation by the analyser trigger another round of feedback, thus initiating a cycle of iteration. This iterative process establishes a self-learning feedback loop that continually refines our investment rules.

 Exhibit 5: Fidelfolio’s Investment Process

These rules are constructed upon the premise set by our investment team—such as the creation of ’13-14-24-20-8- 11-C’ – which emerged when our system was tasked with devising a long-term strategy characterized by low turnover and moderate risk. Our system generates a multitude of rules aligned with the specified criteria and systematically picks out the most promising ones.

Subsequently, our investment team validates these prime rules based on their underlying fundamental rationale, subjecting them to scrutiny and additional assessments. Ultimately, by combining these ‘best’ investment rules created for a particular premise, such as ‘long-term with average risk’, we culminate in a comprehensive and robust investment strategy.

THE WINNING APPROACH – ML WITH HUMAN WISDOM

To solve for the 3 major pitfalls in investment research process, we devised a framework using ML with human intelligence to create transparent investment-rules based on fundamental ratios to create ‘practical’ investment products. ‘Practical’ is underscored by ensuring :

• Interpretability & Transparency – understandable & monitorable by humans and not just another black-box algorithm
• Enabling   ‘Investment’   vs.   trading   –   based   on fundamental analysis for long-term wealth creation 
• Scalability – Low churn to keep it operable at large AUM

By combining ML with human wisdom, FidelFolio addresses the limitations of human biases, limited analytical ability, and human errors. It enables extensive study, pattern identification, understanding of return drivers, elimination of biases and errors, and qualitative analysis. This approach surpasses human-only and machine-only approaches, positioning FidelFolio at the forefront of innovative investment management.

Exhibit 6: ML with human wisdom

Disclaimer: Investment in securities market are subject to market risks. Read all the related documents carefully before investing. Registration granted by SEBI, membership of BASL (in case of IAs) and certification from NISM in no way guarantee performance of the intermediary or provide any assurance of returns to investors. Visit bit.ly/sc-wc for more disclosures.