Siwen Liu

Passionate about quantitative research and data analysis.

• Involved in the internal program regarding fertility prediction as a thesis student. The VAR model and the ordinary linear regression model are constructed based on Google Trends and Twitter data to project fertility.
• Developed a semi-automated fertility keywords generation and selection method based on Natural language processing to avoid time cost and bias by subjective selection.

Utilized Latent Dirichlet Allocation from Gensim to generate fertility related topic according to the preliminary keywords.

• Constructed two SARIMA models and achieved 12 lags(3 years) ahead GDP prediction with the 95% prediction interval.
• Compared and selected the best model with MAE of 94678.57 and RMSE of 250.7 and tested the model differences by Diebold-Mariano test.
• Integrated and constructed VAR model between two time series (GDP and tax receipt) and observed the incremental explanatory power of tax receipt in the prediction of GDP.