A time-series notebook that explores MercadoLibre search interest, compares it with stock-price behaviour and forecasts search traffic with Prophet.
forecasting_net_prophet.ipynb:
- analyses hourly Google search traffic for daily, weekly and annual patterns;
- compares May 2020 traffic with the median monthly total;
- joins search traffic with hourly stock prices;
- measures correlations between lagged search interest, stock volatility and hourly returns; and
- fits a Prophet model and forecasts a further 2,000 hours.
The recorded analysis contains 37,106 hourly search observations from June 2016 to September 2020. May 2020 totalled 38,181 search-interest units against a median monthly total of 35,172.5. Lagged search traffic had weak recorded correlations with stock volatility (-0.149) and hourly returns (0.018). The model components place the highest average search interest around midday and on Tuesdays.
The notebook was written for Google Colab and uses google.colab.files to upload data. Open it in Colab, run the setup cells, then upload the matching CSV from Resources when prompted.
The embedded installation cell reflects an older environment: it installs fbprophet while the notebook imports prophet. No dependency lockfile or currently verified Colab environment is included, so the installation cell may need to be updated before the notebook runs.
The repository includes hourly Google search trends, hourly stock prices and daily revenue data. Their original providers, collection methods, licences and redistribution terms are not recorded.
The search model uses a historical supplied dataset and has no held-out forecast evaluation. Correlation does not establish causation or a trading signal. Recorded notebook outputs show a result from the original environment, not a current reproducibility check.
The optional revenue-forecast section remains unfinished and contains placeholder code. It is not part of the claimed project result.
This project began as the Module 11 challenge in the 2022 Monash University FinTech Bootcamp. The course supplied the notebook structure and datasets. My contribution completed the search-pattern analysis, stock/search comparison, Prophet search forecast and written interpretation.