paper Wagner Brauer Turkey

Uploaded from authorPOINT
Views:
 
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Using Dynamic Forecasting Genetic Programming (DFGP) to Forecast U.S. GDP with Military Expenditure as an Explanatory Variable: 

Using Dynamic Forecasting Genetic Programming (DFGP) to Forecast U.S. GDP with Military Expenditure as an Explanatory Variable Neal Wagner and Jurgen Brauer Augusta State University Augusta, GA, USA www.aug.edu/~sbajmb/ September 2006

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Classic time-series forecasting techniques Exponential smoothing Regression ARIMA Threshold (G)ARCH All suffer from at least two shortcomings: Functional form is investigator-specified Assume constant data-generating process across all segments of a time-series

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming These are troubling shortcomings E.g., Atesoglu (2002) Linear vs non-linear regression Data: 1947:2 to 2000:2 2001:3 and 2001:4 11 September 2001 would a past functional form still hold? would the data generation process be the same as before? What we search for is a method that … automatically selects and self-adjusts both functional form and time-period

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Heuristic methods … … automate the discovery of functional form and permit different segments of a time-series to derive from different data-generation processes These methods include: Neural networks (NN) Evolutionary computation Genetic algorithms (GA) Evolutionary programming (EP) Genetic programming (GP) … but the duration (length) of the time-series is still investigator determined

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming This paper Uses a GP process where even the duration (length) of the time-series is automatically discovered The new method is applied to forecasting U.S. GDP where milex is an explanatory variable The method is compared to a regression-based forecast

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Suppose fig 3 represents historical data segment 2 is current environment valid to forecast the future segment 1 is old environment not relevant for forecasting if both segments are analyzed, the forecast is distorted … … unless human judgment is brought to bear that assigns certain historical data to the analysis if human judgment about the break point in the data series is faulty, the analysis may still contain ‘old’ environment data what is needed is a forecasting method that can automatically determine the correct analysis segment size or 'window,' i.e., the correct number of historical data to be analyzed.

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming By substitution, equation (6) can be derived:

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming …

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Figure 10 OLS results (full sample)

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Figure 11: Actuals and OLS and DFGP forecasts

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Figure 12: d_OLS and d_DFGP forecasts

Dynamic Forecasting Genetic Programming: 

Dynamic Forecasting Genetic Programming Figure 13: loggdp fitted values from regression against time (measured in quarters, t = 1, …, n)