Financial Time Series and Forecasting Assignment -

The goal of this assignment is to build and interpret factor models and to compare a range of models/methods for forecasting, in the context of a dynamic portfolio allocation problem, regarding investment performance; i.e. you will use forecasts from different methods/models to dynamically set optimal asset portfolios, using different rules. The aim is to assess and compare these competing models and methods for investment purposes, including for risk management. Please see the Guidelines document in addition to the following minimum requirements:
Minimum Requirements (not in order)

1. Choose your own data series which should consist of historical prices and returns for 5 assets which could be used to form an investment portfolio. The chosen time series must include data from 2018. Motivate your choice of assets and your chosen sample of data (e.g. data frequency and sample size). Separate your data into in-sample and forecast period samples. Present a sufficient and relevant exploratory data analysis that illustrates the properties of the in-sample data.

2. Build a factor model with at least 2 factors for your 5 asset series justifying your final choice for the number of factors. Describe the factor loadings and factors found. Discuss the fit and appropriateness of the model. Perform at least one factor rotation on the set of factors in your chosen model and discuss whether the rotation has improved the interpretability of the factors. (Hint: see the Matlab documentation for factor rotation by typing "doc rotatefactors" at the Matlab command prompt.)

3. Motivate and discuss quantitative models you will employ for forecasting the asset returns series. You should choose at least 4 different models/methods here for forecasting each asset's return and volatility. At least 2 of these must be GARCH type parametric models, while the others can be naive, adhoc, non-parametric, etc. Include any tests and any diagnostics you applied to choose or adapt/refine these models.

4. Generate moving origin horizon one forecasts of both return and volatility for each observation in your forecast sample for the models in part 3. Assess the forecast accuracy of these models for forecasting both the returns and volatilities of each asset.

5. Asset portfolio returns are given by the weighted sum of the individual asset returns. Motivate, then clearly present at least three (3) different rules or methods for choosing 'optimal' portfolio weights for use in your study. At least one of these methods must involve the concept of risk management, using a quantitative risk measure appropriate for these asset portfolios to choose portfolio weights. You can be creative with your choices, but you must clearly define and justify each weighting method.

6. For each weighting strategy in part 5 use your forecast models to dynamically assign optimal portfolio weights during the forecast sample period (i.e. do this for each combination of strategy and model). Justify your frequency of optimally choosing portfolio weights; use at least two different frequencies of changing weights (e.g. every period, every 5th period). Under each choice you must ‘update' the portfolio weights at least 5 times in the forecast sample. Also justify how often model parameters are re-estimated (a separate decision).

7. Assess and compare the returns and risk measures from all combinations of portfolio strategies and model types. These must be fully and properly discussed, presented, evaluated, compared and interpreted in detail.

8. Include a final discussion that compares the competing models and methods for portfolio allocation on investment and accuracy criteria. Was all this quantitative effort worth it in this case? Discuss.