veronicahurtado / fintech-module11-challenge

Time Series: Make predictions via building machine learning models around time series data

Predict User Traffic for MercadoLibre

Case study

With the aim of driving growth and revenue for Mercado Libre, this tool imports both financial and user data to a Jupyter notebook, and generates visualisations to understand the time series data.

Data sources

• Google hourly search trends: google_hourly_search_trends.csv
• MercadoLibre daily revenue: mercado_daily_revenue.csv
• MercadoLibre stock price: mercado_stock_price.csv

Tasks to support the Analysis of the data

• Visual depictions of seasonality (as measured by Google Search traffic) that are of interest to the company.
• Evaluation of how the company stock price correlates to its Google Search traffic.
• Prophet forecast model that can predict hourly user search traffic.

Technical Environment

This tool utilises the following technologies:

• Pandas Documentation
• hvplot Documentation
• sklearn Documentation
• HoloViews Documentation
• Prophet Documentation
• Matplotlib Documentation

Disclaimer

Please be aware this is an Academic Case Study. The conclusions from this work should not be considered as financial advice.

---

Report

Visual depictions of seasonality

Total search traffic for May 2020: 38181

Overall monthly median search traffic: 35172.5

There is an increase in the Google search trends for May 2020 compared to the overall monthly median value.

Note: The day of the week with Monday=0, Sunday=6

The highest concentration of traffic occurs every day from 10:00 pm to 2:00 am, specially between Sunday evening and Friday morning. Friday and Saturday evenings are less active compare with the rest of the days. This means the company can get a higher ROI by focusing the marketing strategies to the evenings.

It looks like the traffic goes up slightly during the holiday period, but the search trend is not very clear using a heatmap. Other techniques are required to better identify seasonality.

Correlation of Search Traffic to Stock Price Patterns

Based on both graphs it doesn't look like there is any common trend between the Closing price of MercadoLibre stocks and the Search traffic.

There is no significant correlation between the lagged search traffic and the stock volatility. Similarly, there is no correlation between the lagged search traffic and the stock price returns.

Prophet forecast model

Similar to the Heatmap plot, the Prophet model indicates the time with the most search traffic is from 10 pm to 2 am. Tuesday is the day of the week with the most search traffic. The lowest point for the Search Traffic in the years seems to be around October.