InstiFund

Select stocks based on Institutional Investment Trends and Fundamental Analysis

Abstract

Vietnamese open-end funds recently have seen significant growth in the number of people investing in them, especially after the introduction of Fmarket. Their performances are very persuasive, with many funds having a return of 20-30% per year. In this project, I use the VCBF open-end fund's financial portfolio to select stocks. I believe that the fund's financial portfolio can be used as a good indicator for stock selection. Each month, a list of 3 highest scored stocks will be selected based on the fund's financial portfolio and the stocks' financial ratios. The stocks will be held for 1 month and then rebalanced. The performance of the strategy will be compared with the benchmark index (VN-Index) to evaluate its effectiveness. The results show that the strategy outperforms the benchmark index with a ROI of 42.87% and a maximum drawdown of -15.35% in the out-of-sample period.

Introduction

Vietnamese open-end funds have recently gained popularity among investors, particularly following the launch of Fmarket. Many funds report annual returns of 20-30%, making them attractive investment vehicles. However, retail investors often face challenges in selecting individual stocks due to the complexity of financial analysis and market dynamics.

This project aims to address this challenge by leveraging the VCBF open-end fund's financial portfolio as a key indicator for stock selection. The hypothesis is that stocks held by the fund represent high-quality investments, and by systematically selecting the best-performing stocks based on financial ratios, it is possible to generate superior returns compared to the market benchmark (VN-Index).

The methodology involves a monthly selection of the top three stocks with the highest scores, determined through a combination of institutional and financial metrics. These stocks are held for one month before rebalancing. The strategy is backtested using historical data, and an optimization process is conducted using Optuna to fine-tune parameters for maximum performance.

Results indicate that the strategy achieves an optimized annual return of 59.52% with a maximum drawdown of -11.14% and a Sharpe ratio of 2.08 during the in-sample period. For the out-of-sample period, the method maintains strong performance with a 42.87% annual return, -15.35% maximum drawdown, and a Sharpe ratio of 1.74, consistently outperforming the VN-Index.

These findings suggest that incorporating institutional investment trends into a systematic trading strategy can be a viable approach for retail investors seeking to enhance portfolio returns.

Trading (Algorithm) Hypotheses

In addition to smart-beta strategy that use quantitative factors to select stocks, I also use the VCBF open-end fund's financial portfolio to select stocks. I believe that:

• Open-end funds’ decision based on long-term growth possibility.
• We can aggregate financial portfolios of open-end funds to make our decisions.
• The reported decisions of the last quarter show the funds’ vision of the market in the next several quarters.

Data

• Data source: Algotrade internship database, VCBF, VCI using Vnstock library
• Data type
• Data period: from 2022-01-01 to 2025-01-31
• Both input and output data will be stored in a specified data path. The default data path is data folder in the root directory of the project. You can change the data path by changing the DATA_PATH variable in the .env file.
• Data format: mainly CSV files, JSON files are used to store evaluation results and hyperparameters.

Data collection

VCBF open-end fund financial portfolio

• The VCBF open-end fund's financial portfolio is collected from VCBF website using web scraping. I download excel files from crawled links and extract necessary data to CSV files.
• The data is collected using the src.crawler.vcbf_crawler module. The files are stored in <DATA_PATH>/VCBF/downloaded/.
• Due to VCBF only started publishing report in excel format since late 2022, I only collect data from 2022-11-01 to 2025-01-31.
• The retrieved list of stocks are used to select stocks, which means other stocks that are not in the list will be ignored in the strategy.

Daily price, quantity of Vietnamese stocks

• The daily price and quantity of Vietnamese stocks are collected from Algotrade internship database using SQL queries.
• The data is collected using the src.crawler.stocks_crawler module.
• The data is stored in the <DATA_PATH>/daily_data.csv file.
• By default, the data is collected from 2022-01-01 to 2025-01-31. You can change the time range by changing the start_date and end_date parameters of data_collection in the config/config.yaml file.

Stocks financial portfolio data

• Financial information of the stocks is collected using Vnstock library, which use VCI data source to return financial information of the stocks.
• The data is collected using the src.crawler.stocks_crawler module.
• The data is stored in the <DATA_PATH>/financial_data.csv file.
• The time range is the same as the daily price and quantity of Vietnamese stocks.

Data Processing

Extract VCBF open-end fund financial portfolio

• The downloaded excel files are extracted to CSV files. The results are stored in <DATA_PATH>/VCBF/extracted/ folder.
• These CSV files are then merged into a single CSV file and stored in <DATA_PATH>/VCBF/fund_portfolios.csv file.

Monthly scores preprocessing

• To speed up backtest process, I preprocess the financial scores and institutional scores into a single CSV file named <DATA_PATH>/monthly_scores.csv file.
• The preprocessing process is done using the src.preprocess module.
• Details of the scores calculation are described in the Backtesting section below.

Implementation

Environment Setup

1. Set up python virtual environment

python -m venv venv
source venv/bin/activate # for Linux/MacOS
.\venv\Scripts\activate.bat # for Windows command line
.\venv\Scripts\Activate.ps1 # for Windows PowerShell

2. Install the required packages

pip install -r requirements.txt

3. (OPTIONAL) Create .env file in the root directory of the project and fill in the required information. The .env file is used to store environment variables that are used in the project. The following is an example of a .env file:

DB_NAME=<database name>
DB_USER=<database user name>
DB_PASSWORD=<database password>
DB_HOST=<host name or IP address>
DB_PORT=<database port>
DATA_PATH=<path to the data folder of your choice, if not specified, the default is `data`>

The DATA_PATH variable is used to specify the path to the data folder where the input data is stored. The other variables are used to connect to the database.

Note: Skip this step if you decide to use the provided data files on Google Drive (option 1).

Data Collection

Option 1. Download from Google Drive

You can skip this step by download the necessary data files from Google Drive and put them in the data folder.

• The data files are stored in the data folder. The folder structure is as follows:

data 
├── VCBF
│   └── fund_portfolios.csv
├── financial_data.csv
├── daily_data.csv

Option 2. Run codes to collect data

1. Retrieve VCBF open-end fund financial portfolio In the root directory of the project, run the following command. The process will take about 7-10 minutes to finish.

python -m src.crawler.vcbf_crawler

The result will be stored in the <DATA_PATH>/VCBF/fund_portfolios.csv file with other downloaded files from VCBF's website. The data is stored with the following format:

Fund Code,Date,Category,Quantity,Market Price,Value,Total Asset Ratio
VCBF-BCF,2023-01-01,ACB,504488.0,26050.0,13141912400.0,0.0352110336932618
...

2. Retrieve daily price, quantity and financial data of Vietnamese stocks If you have access to Algotrade internship database, you can use the following command to retrieve the data.

python -m src.crawler.stocks_crawler 

The data time range is configured in the config/config.yaml file. The default time range is from 2022-01-01 to 2025-01-31. You can adjust the time range by changing the start_date and end_date parameters of data_collection in the config/config.yaml file.

The daily stock price results will be stored in the <DATA_PATH>/daily_data.csv file. The data is stored with the following format:

datetime,tickersymbol,price,quantity
2024-12-31,VIC,40.55,1784400
...

Stocks financial portfolio data will be stored in the <DATA_PATH>/financial_data.csv file. The data is stored with the following format:

Revenue,year,quarter,Cash,Liabilities,P/E,ROE,Financial Leverage,Debt/Equity,tickersymbol
9053845000000,2022,1,6281931000000,480433095000000,7.7889284027,0.236563115,10.966802399327864,9.9668023993,ACB
...

Data processing

Now we preprocess the financial data and institutional data to prepare for backtesting step.

python -m src.preprocess

The result will be stored in the <DATA_PATH>/monthly_scores.csv file. The data is stored with the following format:

symbol,fund_net_buying,number_fund_holdings,net_fund_change,roe,debt_to_equity,revenue_growth,pe,month,year
ACB,0.18949771689497716,3.0,3.0,0.2649167736,9.401935188,15.362221323862574,6.1561819167,2,2023
...

In-sample Backtesting

• To init parameters for the first run, access config/config.yaml file and adjust the default_backtest_params.
• For this project, we use data of the period from 2023-02-01 to 2024-01-31 as the in-sample period. These can be adjusted in config/config.yaml.
• --name argument must match the configuration name in config/config.yaml file, where each name has start_date and end_date parameters. The default name is in_sample. This name will also be used to save in the result folder.

python -m src.backtest --name in_sample

The result will be stored in the <DATA_PATH>/backtest/in_sample folder.

Optimization

Run this command to start the optimization process. You can adjust the random seed by editing random_seed in the config/config.yaml file. The default random seed is 42.

python -m src.optimize --n_trials 5000 

The optimization result will be stored in the <DATA_PATH>/optimization folder. And the optimized in-sample backtest result will be stored in the <DATA_PATH>/backtest/optimized_in_sample folder. This process will take about 1-2 hours to finish on a standard laptop. You can skip this step by copying the best_params.json file to the <DATA_PATH>/optimization folder.

After that, you can run the in-sample backtesting process again with the optimized parameters.

python -m src.backtest --name optimized_in_sample

The result will be stored in the <DATA_PATH>/backtest/optimized_in_sample folder.

Out-of-sample Backtesting

Run this command to start the out-of-sample backtesting process.

python -m src.backtest --name out_sample

The result will be stored in the <DATA_PATH>/backtest/out_sample folder.

Configurations

• The configuration file is stored in the config/config.yaml file. You can adjust the parameters in this file to change the behavior of the algorithm.
• Read the config/README.md file for more details about the configuration file.

In-sample Backtesting

Stock Selection

Score of a stock is calculated by the following formula:

$$\text{score} = x_1 \cdot \text{institutional\_score} + x_2 \cdot \text{financial\_score}$$

The institutional score is calculated by the following formula:

$$\text{institutional\_score} = x_3 \cdot \text{fund\_net\_buying} + x_4 \cdot \text{number\_fund\_holdings} + x_5 \cdot \text{net\_fund\_change}$$

Where:

• fund_net_buying is the net buying of the fund in the month.
• number_fund_holdings is the number of funds holding the stock.
• net_fund_change is the net change of the fund in the month.

The financial score is calculated by the following formula:

$$\text{financial\_score} = x_6 \cdot \text{roe}+ x_7 \cdot \text{debt\_to\_equity} + x_8 \cdot \text{revenue\_growth} + x_9 \cdot \text{pe}$$

Where:

• roe is the return on equity.
• debt_to_equity is the debt to equity ratio.
• revenue_growth is the revenue growth rate.
• pe is the price to earnings ratio.

The weights $x_1, x_2, \dots, x_9$ are the parameters that we want to optimize.

Rebalancing and Risk Management

• Portfolio is rebalanced every month, at 20th of the month, when the new quarterly financial report is published, and new monthly open-end fund report is published (usually on the 8th of the month).
• Rebalancing process takes 2 days.
• On the first day, we sell all stocks in the portfolio to achieve 100% cash.
• On the second day, we execute these steps:
  1. Scoring and ranking the stocks based on the mentioned formula, and select the top 3 stocks with the highest score.
  2. Depend on the scores, we distribute the amount of cash to buy each stocks. There are 3 options: equal, linear, softmax, this will also be optimized in the optimization step.
  3. Buy the stocks with the allocated cash amount.
• To manage the risk, we set a trailing stop loss and a take profit. These parameters are also optimized in the optimization step.

Evaluation Metrics

• Backtesting results are stored in the <DATA_PATH>/backtest/ folder.
• Used metrics to compare with VNINDEX are:
  • Return on Investment (ROI)
  • Maximum drawdown (MDD)
  • Sharpe ratio (SR)
  • Sortino ratio (SoR)
  • Calmar ratio (CR)
  • Votality (Vol)
  • Max Time to Recover from a drawdown (MTR) in days
• Other metrics:
  • Cumulative annual growth rate (CAGR)
  • Win rate: the percentage of winning trades over total trades.

Parameters

The default parameters are stored in the config/config.yaml file.

default_backtest_params:
  institutional_weight: 0.6
  take_profit: 0.325
  trailing_stop_loss: 0.5
  fund_net_buying: 0.35
  number_fund_holdings: 0.45
  roe: 0.3
  revenue_growth: 0.1
  pe: 0.4
  stock_weight_option: "equal"

• net_fund_change is set so as to sum up to 1 with the other parameters in the institutional score.
• debt_to_equity is set so as to sum up to 1 with the other parameters in the financial score.

In-sample Backtesting Result

• Table of the backtesting results, compare with VNINDEX benchmark from 2023-02-01 to 2024-01-31.

	ROI	MDD	SR	SoR	CR	Vol	MTR
In-sample	12.88%	-12.13%	0.47	0.62	1.07	19.03%	224
VNINDEX	8.24%	-17.44%	0.26	0.31	0.47	16.03%	223

• Other metrics:
  • CAGR: 12.97%
  • Win rate: 47.22%
• Cummulative return plot of initial backtesting results, along with VNINDEX benchmark.
• Maximum drawdown
• Daily return
• Cash flow

Optimization

• Library: Optuna
• Sampler: TPESampler
• Number of trials: 5000
• Optimization objective:

$$0.8 \cdot \text{SR\_score} + 0.2 \cdot \text{MDD\_score}$$

Where:

• SR_score varies from 0 to 1 as SR varies from 0 to 3.0.

• MDD_score varies from 0 to 1 as MDD varies from -20% to -5%.

• Check out optimize.py for more details.

• Optimization history

• Parameter importance plot

• Optimized parameters: can be found in best_params.json file.

{
    "trailing_stop_loss": 0.3,
    "stock_weight_option": "equal",
    "take_profit": 0.375,
    "institutional_weight": 0.47500000000000003,
    "fund_net_buying": 0.22500000000000003,
    "number_fund_holdings": 0.425,
    "roe": 0.05,
    "revenue_growth": 0.47500000000000003,
    "pe": 0.22500000000000003
}

Optimization Result

• Table of the optimized backtesting results, compare with VNINDEX benchmark from 2023-02-01 to 2024-01-31.

	ROI	MDD	SR	SoR	CR	Vol	MTR
Optimized In-sample	59.52%	-11.14%	2.08	2.88	5.39	21.02%	186
VNINDEX	8.24%	-17.44%	0.26	0.31	0.47	16.03%	223

• Other metrics:
  • CAGR: 59.98%
  • Win rate: 61.11%
• Cummulative return plot of initial backtesting results, along with VNINDEX benchmark.
• Maximum drawdown
• Daily return
• Cash flow

Out-of-sample Backtesting

• The out-of-sample period is from 2024-02-01 to 2025-01-31.
• The out-of-sample backtesting process is the same as the in-sample backtesting process, but we use the optimized parameters from the optimization step.

Out-of-sample Backtesting Result

• Table of the backtesting results, compare with VNINDEX benchmark from 2024-02-01 to 2025-01-31.

	ROI	MDD	SR	SoR	CR	Vol	MTR
Out-of-sample	42.87%	-15.35%	1.74	2.39	2.86	19.43%	194
VNINDEX	8.01%	-8.94%	0.27	0.34	0.90	13.70%	223

• Other metrics:
  • CAGR: 44.05%
  • Win rate: 66.67%
• Cummulative return plot of initial backtesting results, along with VNINDEX benchmark.
• Maximum drawdown
• Daily return
• Cash flow