แจ้งเตือน

FINNOMENA Best-In-Class Methodology White Paper

Once the Best not always the Best

Abstract 

     There is a variety of ways to invest in mutual funds. Two of the main traditional ways that most investors rely on are historical returns and analyst’s recommendations. Those mentioned investing ways are not always meeting investor’s expected returns in long terms. This experiment is aiming to advise a set of mutual funds in different classes through processing machine learning. Selected mutual funds are expected to generate the highest 6-month forward returns relative to total return SET index and peers in each asset by 12.83 %

1. Introduction

     Relying on traditional investing ways seems to be impossible for investors to meet their financial goals amid the current market environment with high volatility and unexpected events. Over the past few years, investors are more sophisticated and experienced in the financial markets, but it is still not enough to survive and meet their financial goals, especially in this current market situation.   

     Portfolio Finnomena Best-In-Class (BIC) can help investors select the best mutual in each category of asset class through processing the Random Forests Model (RFM), one of the Machine Learning applications. RFM can screen and select the best mutual funds in different five categories that are expected to generate the highest risk-adjusted returns in six months forward compared to the benchmark and their peers. The inputs of processing RFM model include economic data, mutual fund’s return seasonality, applied-mathematic data, technical indicator, and the views from fund managers and analysts. RFM model will run all of the variables and determine which one is statistically significant to forecast BIC’s return. This will benefit us to improve the efficiency of BICs model in the future.

2.Data and Methodology

2.1 Data Source

     The inputs of processing RFM include daily 20-year historical returns of Equity Thai Active Large-Cap, technical indicators, views from fund managers and analysts, applied-mathematic data, and NAV of each mutual fund. RFM model can determine which variables are statistically significant to forecast returns of BICs. Hence, it can select the inputs that are significant to the model, which will enhance the accuracy of our models to select the best possible mutual funds in each asset class. The criteria of input selection are that it is widely known by local investors and enough number of mutual funds in Equity Thai Active Large-Cap.

2.2 Portfolio Construction Method

     The portfolio will select the top three mutual funds with the highest score through running BICs models every six months

2.3 Experiment Setup

     In this experiment, we set the definition of the best portfolio to be the one that generates the maximum alpha in a given period.

α = ReturnPortfolio – ReturnBenchmark

However, instead of using market return as a benchmark, in this experiment, we use the average return of mutual funds within the same category served as benchmark return to enhance our precision. Next, we evaluate the strategy based on the alpha that the portfolio generated at the end of each of the given periods. The invested period is set to be 6 months, meaning that each of the constructed portfolios will be invested for 6 months and the generated return during that period will be evaluated against the mutual funds’ return in the same category at the end of the 6th month

3. Model Architecture

     Apart from quantitative data, this experiment also relies on views from our investment team who have experienced and specialized in selecting mutual funds. With the combination of quantitative and qualitative data, it enhances the confidence level of the model to select the mutual funds that can achieve the 6-month highest possible return relative to peers in each category. 

3.1 Random forests Model [1]

     Significant improvements in classification accuracy have resulted from growing an ensemble of trees and letting them vote for the most popular class. To grow these ensembles, often random vectors are generated that govern the growth of each tree in the ensemble. (Leo Breiman, 2001) 

Figure 1. Random Forests Model

3.2 Best In Class Model

     BIC portfolio with RFM can select a set of mutual funds in each category, which is expected to achieve the highest possible alpha (the difference of returns between selected mutual funds and peers in the same category. Through processing Ensemble Learning, the model creates additional trees, and then the model will select the tree with the largest number of votes from every tree. Sequentially, the model will rank the score of mutual funds based on results of RFM to create BICs portfolios 


Figure 2. Best-In-Class architecture

4. Best In Class Portfolio

     After we got the score for each mutual fund from the model in Sec. xx, we then allocate the 3 highest scored mutual funds at 40%, 30%, and 30% to form a Best in Class Portfolio based on the chosen category. The portfolio is then adjusted and rebalanced every 6 months to ensure the performance as we have shown in Sec. xx, that the historically performed mutual funds do not necessarily excel in the future. Please note that each Best In Class Portfolio used a specific model that was trained on the delegated category.

4.1 Best In Class Experiment Results 

     From the result of the experiment, the selected mutual fund (BICs) in Equity Large Cap outperformed the peer’s average 2-year returns. Apart from the returns, selected mutual funds outperform peers in metrics such as alpha ratio by 12.83% and lower maximum drawdown by 4.56 % 

4.2 Trading Simulation 

     Performance evaluation of the BICS model rebalanced the portfolio every 6 months under the real-world scenario while holding the switching period to be T+3, transaction fee to be at 0.5% for both buy and sell as well as 1% for each complete transaction. From the chart below, it represents the returns of BICs performance relative to benchmark SET TRI


Figure 3.
A Balanced BIC Portfolio Strategy
40% 1st pick, 30% 2nd pick, 30% 3rd pick

Bull Market Environment 

     During the bull market environment, given rebalancing the portfolio every six months during 2017-2018, BICs portfolio’s net-of-fee return is 28.14 % while the benchmark SET TRI return is 12.14 %. The BICs model has a positive alpha ratio of 16 %. 


Figure 4.
A BIC Portfolio Strategy in Bull Market

Bear Market Environment

     During the bear market environment ranging from 2018-2019 , BICs still outperformed SET TRI return. it has a smaller maximum drawdown of -11.24 % while SET TRI has -15.8 % The performance comparison has been shown below.    


Figure 5.
A BIC Portfolio Strategy in Bear Market


Figure 6.
Historical returns from Strategies

5. Conclusion 

     The experiment proposes an alternative way to select mutual funds through processing Random Forests Model, one of the machine learning model types. Based on the back-testing result, it has been found that the BICs portfolio significantly outperforms the benchmark in different market scenarios. Apart from quantitative metrics used in the model, we also evaluate the fund manager’s performance over time as a part of BICs’ score. Overall, we found this model to be well-suited for assisting individual investors to construct their portfolio

References

  1. LEO BREIMAN., Machine Learning, 45, 5–32, 2001 c 2001 Kluwer Academic Publishers. Manufactured in The Netherlands.
  2. Thomas G Dietterich., Ensemble Methods in Machine Learning, 2000, Oregon State University Corvallis Oregon USA.