Real Life Problem This one is a tough one and some crowd sourcing seems like a good way to get some feedback. I am trying to determine the effect of Non-Farm Payroll surprises on a subsector of the economy. Furthermore, I want to forecast the effect on the subsector using my own surprise estimates for an upcoming non-farm payroll event. I want to take into account the subsector's movements due to its correlation with Equities and Rates.
Stats Problem I am trying to determine the effect of an exogenous shock on a time series (so series of shocks over time of varying magnitude) taking into account the time series correlation with two other time series. I want to then forecast the result of an upcoming shock on the time series.
Current Solution Ideas Listed in order of 'complexity'
- Linear regression of panel data focusing on day of changes in time series NFP in relation to the shock
- MultiLinear regression of panel data focusing on day of changes in time series NFP in relation to the shock taking into account Stock and Rates movements
- MultiLinear regression of time series data focusing on day of changes in time series NFP in relation to the shock taking into account Stock and Rates movements. Shock would be dummy crossed with magnitude.
- Autoregressive distributed lag model using BIC/AIC to determine lag numbers and checking inclusion of stock and rates timeseries with Granger causality. Shock would be dummy crossed with magnitude.
- Other? What am I missing?
The dataset is somewhat stationary, the timeframe is usually limited (3-5 years, with 36-60 shocks) so trend changes are minimal
Note: Try them all and see best fit is not an answer, that's just data mining ;)
