Navigating Investments: Building Diverse Portfolios with Tableau Dashboard

By: Julia Tsai, Ke-Hsuan Chu, Suzanne Pang

View the dashboard: Tableau Dashboard

In the realm of asset management sales where there are countless investments to choose from, investment strategies can be optimized using the crossroads of data science and strategic visualization. Using portfolio optimization, we created a dashboard to help asset management professionals in guiding clients toward resilient investment strategies centered around the company’s products.

Navigating the Investment Dashboard

Funds Overview

We took a look at our our funds’ total assets ranking (Table A) , where the funds’ focus are on (Table B), the funds’ returns over month as well as their returns and risks (Table C). In Table C, a fund that has high returns and low risk would appear on the upper left side of the graph.

Table A

Table B

Table C

Understanding Client’s Portfolio

From this point, our goal is to understand the client’s portfolio interests and investment aspirations to construct a personalized portfolio that ensures the lowest risks.

For the current dashboard, we have simulated three different portfolios constructed by our clients. Subsequent charts will analyze and visually present data based on these investment portfolios. However, we plan to develop an interactive real-time panel in the future. This will allow clients to simulate and build investment portfolios directly within the dashboard. Subsequent charts will then be linked to Python code for immediate analysis and presentation.

The graphs help the client in gaining a stronger understanding on the new portfolio.

Graph A displays the comparison of the portfolio old vs new through different periods. Through this graph, we can see that, especially in the recent period, the new portfolio provides lower risks compared to the old portfolio.

Graph A

Graph B shows the returns and risks of the recommended funds displayed in the new portfolio on the top and it displays the contrast between the old and new portfolio in terms of returns and risks on the bottom.

Graph B

Fund Similarity within Portfolio

The final section of the dashboard provides clients with an overview of the similarity between funds through a cosine similarity heatmap located at the top of the page.

At the bottom of the dashboard, the new portfolio’s funds are presented alongside their principal component scores. Consistent with the earlier graph, there is no notable similarity between any two funds in the portfolio.

Quantifying Success within New Portfolio

By aligning strategic vision with quantitative insights, our portfolio optimization tool works to serve as a visualization for asset management teams to navigate the intricate landscape of modern asset management with confidence. When comparing the old and new portfolios, we found that although returns are similar for both, the risks are significantly lower in the new portfolio. The investment dashboard tool is a valuable resource for novice investors seeking passive income, aiding them in identifying low-risk funds for investment opportunities. This tool crafts client portfolios that stand resilient in the face of market complexities.

---

Data Analysis

Data Preparation

To begin, we first cleansed the data, filled in missing values, and converted string values to object values to have a deeper exploration of inherent patterns and relationships.

K Means & PCA Analysis

We used the elbow method and determined that we needed to create 2 clusters.

We used PCA analysis to reduce dimensions and choose 6 principle components (Graph A). These components, chosen for their ability to explain over 77.61% of the variance, guide asset managers toward streamlined and insightful analyses of the company’s product landscape.

Graph A

Graph B show the clusters relationship according to various principal components.

Graph B

Graph C shows the features of each principal component and reveals the main factors among each of them. The loadings of each principal component are shown in the following figure, and we have identified funds that are more closely related to each principal component by using twice the standard deviation.

It is worth to mention that:•⁠ ⁠Risk performance of funds are highly positively correlated with PC1•⁠ ⁠PC2 has a higher positively correlation with the returns of funds, especially medium and long term return rates•⁠ ⁠The funds’ rating class, strategy, and average dividend yield are negatively correlated with PC3. From these results, we can infer that PC3 is more closely related to the fund’s investment decisions.

Graph C

Fund Selection Process

We removed funds by examining their distribution on the PCA scores scatterplot. Using the Density-Based Spatial Clustering of Applications with Noise method, we set the minimum sample point to 3, and were able to separate real clusters from random noise. We also ensured that the funds that were selected to add to the portfolio exhibit a cosine similarity below 0.86 (with a 60-degree angle). This step showcases the tangible impact of choosing funds with low similarity — an approach designed to diversify risks and fortify investment resilience.

This method enabled us to establish a criteria-driven selection of funds, aligning with our objective to refine asset portfolio selection. Asset managers can now strategically eliminate funds with high similarity. The criteria emphasizes low cosine similarity and strategic positions on PCA scores scatter plots, transforming the selection of new funds into a cross-section of diversification and risk mitigation.

Upon finalizing the fund selection process, we repeatedly computed the returns and risks associated with investment portfolios containing the chosen funds meeting our selection criteria. The portfolio with the lowest observed risks over the past month was then identified and retained as the recommended investment portfolio.