DATA MINING TECHNIQUES FOR PORTFOLIO BUILDING

Анотація

Data Mining has revolutionized portfolio building and stock forecasting by providing advanced
methods to analyse complex financial data. Techniques like Deep Neural Networks (DNNs)
and Recurrent Neural Networks (RNNs) enable accurate predictions by identifying patterns in historical
stock prices and market trends. These models enhance decision-making by reducing forecasting
errors and improving portfolio performance. Leveraging their ability to process large datasets
and adapt to dynamic financial environments, deep learning plays a crucial role in optimizing
investment strategies and mitigating risks.
According to Markowitz's theory, the expected return of a portfolio is determined using a
formula that incorporates multiple factors and their interconnections:
Rp= iRiwi,#1
where Rp is the return on the portfolio, Ri - return on the asset, wi - share of the asset in the
portfolio.
The first criterion is the expected return function and second criterion is to minimize the total
portfolio risk as follows:
f1=Rp →max, f2= p→min, #(2) p=p2,
where p-the portfolio's expected risk, or standard deviation.
An optimization problem to find the optimal stock portfolio [1]:
W=α*wT*cov*w1-α*i=1nRi*wi →min {i=1nwi=1 wi≥0,01 i=1nRi*wi>0 wT*cov*w≥0 ,#3
This study will utilize two types of neural networks: DNN and RNN. The Deep Neural Network
(DNN) architecture will comprise four layers: Dense(150), Dense(100), Dense(50), and
Dense(1), with each layer employing the ReLU activation function to enhance non-linear feature
learning. Similarly, the Recurrent Neural Network (RNN) architecture will consist of two SimpleRNN
layers, each containing 40 neurons, followed by a Dense(1) output layer. Consistent with
the DNN, all layers within the RNN will utilize the ReLU activation function to ensure effective
data processing and feature extraction [2, 3].
Upon obtaining the results, a comparative analysis of the predictive accuracy between the two
models will be conducted. The findings will be summarized and presented in Table 1.
After obtaining the forecasting results, optimal portfolios will be constructed and compared
against the actual portfolio to evaluate performance and alignment with real-world outcomes. The
results will be presented in Table 2 and Table 3 for detailed analysis and comparison: