Time-Series-Forecasting

In this report, we will work with two datasets. The first one gives us the date of a property’s sale, the price, the property type (unit or house) and finally the number of bedrooms (1, 2, 3, 4 or 5). The second dataset transforms the raw sales data (first dataset), by re-sampling it at quarterly intervals with median price aggregation and step 4 moving average smoothing. The date range of these sales is from 09/2007 to 09/2019. To better understand our data, we will first conduct a Exploratory Data Analysis (EDA). Then we will focus on the goal of the report which is to forecast for the 8 future quarters, the sale price for each property type and number of bedroom. In order to forecast, we will build and compare two types of models: one model from the traditional multi-variate forecasting models from the VARMAX family and one model from Recurrent Neural Network (RNN) such as the Long Short-Term Memory (LSTM) model.