setup
1
Introduction
2
Curriculum
3
Qualitative Forecasting Methods
3.1
Judgmental Forecasting and Forecast Adjustments
3.2
Articles
3.2.1
A new product growth model for consumer durables,” Management Science
3.2.2
Sales Forecasts for Existing Consumer Products
3.2.3
Judgmental forecasting A review of progress over the last 25 years, International Journal of Forecasting
4
Exploring Patterns and Forecasting Techniques
4.1
The forecasting process
4.2
Data Patterns and terminology
4.2.1
Terminology
4.2.1.1
Autocorrelation
4.2.1.2
Random vs. correlated data
4.2.1.3
Stationary vs. non stationary data
4.3
Data Patterns and Model Selection
4.3.1
Forecast Methods and Horizons
4.4
Exercises
4.4.1
p. 92 HW Problem 8
4.4.1.1
Moving Averages
4.4.1.1.1
Regular MA
4.4.1.1.2
MA using loop
4.4.1.2
Exponential moving averages + Holts and Winters
4.4.2
p. 93 HW Problem 9-10
4.4.3
CO2 and Sales Data
4.4.4
Case 6 oo. 108-111 HW
5
Simple and Multiple Linear Regression
5.1
Simple Linear Regression
5.1.1
Assumptions
5.1.1.1
Serial correlation (checking for independent residuals):
5.1.1.2
Heteroskedasticity:
5.1.2
Forecasting with a linear trend
5.1.3
Exercises
5.1.3.1
Problems 5 pp. 209
5.1.3.2
Problem 11 p. 212 HW
5.1.3.3
Cases 2 HW
5.1.3.4
Case 3 from HW
5.1.3.5
Detrending thorugh regression: CO2
5.2
Multiple Linear Regression
5.2.1
Multicollinearity
5.2.2
Serial correlation and omitted variables
5.2.3
Selection criteria
6
Time-Series Decomposition and Regression with Time-Series Data
6.1
Time Series and Their Components (HW)
6.1.1
Additional on trend
6.1.2
Additional on seasonal pattern
6.1.3
Cyclical and Irregular Variations
6.2
Methods of decomposing
6.3
Regression with time series data
6.4
The success criteria and process
6.4.1
Success Criteria
6.4.2
The Process
6.4.2.1
Desaesonalizing
6.4.2.2
Long-term trend
6.4.2.3
Cyclical Component
6.4.2.4
Time-Series decomposition forecast
6.4.3
Autocorellation
6.5
Exercises
6.5.1
Decomposition + Forecast Exercise
6.5.1.1
a. Decomposing the time series
6.5.1.2
b. Retrieving components
6.5.1.3
c. Interpreting the resutls
6.5.1.4
d. Forecasting using the components
6.5.1.4.1
Fitting the model
6.5.1.4.2
Forecasting using the composition
6.5.2
Decomposition with cycles + Forecasting
6.5.2.1
1. Finding CMA (deseasonalizing)
6.5.2.2
2. Finding SF
6.5.2.3
3. Identifying Trend
6.5.2.4
4. Finding detrended and deseasonalized data
6.5.2.5
5. Cycle identifying factor
6.5.3
Time series regression
6.5.4
Alomega Food Stores, case 6 p. 166 + case 7 p. 348 (Another example)
6.5.4.1
What might Jackson Tilson say about the forecasts?
6.5.4.2
P.348 Case 7 (Alomega Food Stores)
6.5.5
Monthly Sales Data
6.5.6
Own test - Monthly Sales Data decomposition + forecast
7
ARIMA Models + ADL and Box-Jenkins Methodology
7.1
ARIMA
7.1.1
Elaborating on AR models
7.1.2
Elaborating on MA models
7.1.3
Elaborating on Integration models
7.2
Model Building Process
7.3
Advantages and disadvantages for ARIMA models
7.4
Dynamic Forecasting
7.4.1
ADL
7.4.2
Vector autoregressive (VAR)
7.4.2.1
Selection criteria
7.4.2.2
Process
7.4.2.3
Analyzing response to shocks in other variables - Cholesky Ordering
7.5
Exercises - ARIMA
7.5.1
IBM stock, problem 12 p 405
7.5.2
Demand data, problem 7 p 403
7.5.3
Closing stock quotations, problem 13 p 409
7.5.4
HW: Case 1 page 413-414 (q1-3)
7.5.5
HW: Case 4 page 417-419
7.5.6
Sales data seasonal
7.5.7
In class assignment
7.6
Exercises - ADL
7.6.1
GDP and CO2 levels
7.6.2
Short- and long-term interest rates, exchange rates
7.6.2.1
Loading data, combining time series and assessment of stationarity
7.6.2.2
Vector AutoRegressive Model (VAR)
7.6.2.2.1
1. Determine order of lags to be included + model estimation
7.6.2.2.2
2. Model diagnostics
7.6.2.2.3
3. Response to shocks in the variables
7.6.2.2.3.1
Exogenious vs. endogenious
7.6.2.2.4
4. Forecasting with ADL
7.6.2.2.4.1
Step 1 - Data partition
7.6.2.2.4.2
Step 2 - Select the order of VAR
7.6.2.2.4.3
Step 3 - Fit the model + check residuals
7.6.2.2.4.4
Step 4 - Make the forecast
7.6.2.2.4.5
Step 5 - Assessing accuracy
8
Non Stationary Time-series
8.1
Unit Roots
8.1.1
Augmented Dickey-Fuller (ADF) test
8.2
Spurious Regression
8.3
Cointegration
8.3.1
Checking for cointegration
8.4
Exercises
8.4.1
Dairy Data
8.4.1.1
Unit Root testing - Augmented Dickey-Fuller (ADF)
8.4.1.2
Cointegration
8.4.1.2.1
Graphical inspection of the data
8.4.1.2.2
Test for cointegration
9
Combining Forecast Result and Forecast Evaluation
9.1
Why, when and how to combine forecasts
9.1.1
Why do we combine forecasts?
9.1.2
When to combine forecasts?
9.1.3
How to combine forecasts?
9.2
Exercises
9.2.1
Air passengers
9.2.1.1
Producing forecasts + combined forecast
9.2.1.1.1
1. Constructing forecasts
9.2.1.1.1.1
a. Forecast 1 - ARIMA
9.2.1.1.1.2
a. Forecast 2 - HoltWinters
9.2.1.1.2
2. Diebold-Mariano test
9.2.1.1.3
3. Combining the forecasts
9.2.1.1.3.1
a. Nelson Combination Method - her code is put here, but not further explained
9.2.1.1.3.2
b. Granger-Ramanathan
10
Exam cases
10.1
Exam 2018
10.1.1
Case 1
10.1.1.1
Q1
10.1.1.2
Q2
10.1.1.2.1
Forecasting using ARIMA
10.1.1.2.2
Forecasting with simple moving averages 1991 - 2000
10.1.1.2.2.1
1. Using ma()
10.1.1.2.2.2
2. Using excel
10.1.1.2.2.3
3. Using a loop
10.1.1.3
Forecasting 2001 - 2018
10.1.1.3.1
Using ma (loop method)
10.1.1.3.2
Using ARIMA (RW + drift)
10.1.2
Q3
10.1.2.1
Diagnostics
10.1.2.1.1
Check for heteroskedasticity
10.1.2.1.2
Check for independent residuals
10.1.2.1.3
Checking residuals for normality
10.1.2.1.4
Conclusion
10.1.3
Q4
10.1.3.1
Test without integration
10.1.3.2
Test with first order integration
10.1.4
Q5
10.1.5
Q6
10.1.5.1
Vector AutoRegressive Model (VAR)
10.1.5.1.1
1. Determine order of lags to be included + model estimation
10.1.5.1.2
2. Model diagnostics
10.1.5.1.3
3. Response to shocks in the variables
10.1.5.1.3.1
Exogenious vs. endogenious
10.1.5.1.4
4. Forecasting with ADL
10.1.5.1.4.1
Step 1 - Data partition
10.1.5.1.4.2
Step 2 - Select the order of VAR
10.1.5.1.4.3
Step 3 - Fit the model + check residuals
10.1.5.1.4.4
Step 4 - Make the forecast
10.1.5.1.4.5
Step 5 - Assessing accuracy
10.1.6
Q7
10.1.7
Case 2
10.1.7.1
Q1
10.1.7.2
Q2
10.1.7.3
Q3
10.1.7.4
Q4
10.1.7.5
Q5
10.1.7.6
Q6
10.1.7.6.1
Combined model
10.2
Nordpool Case
10.2.1
1.What is the key variable of interest and which ones are the variables that are of secondary interest?
10.2.2
2 What is the relationship between system price and buy/sell sides?
10.2.3
3 What models should we use to build an optimal forecast for system price?
10.2.4
4 What is the expected behavior of system price in the next 25 periods?
10.2.5
5 How does the future relationship of system price with the other variables evolve?
10.2.6
6 What can we infer from this relationship, and how can we give investment recommendation based on that?
11
Communicating Technical Findings
12
Methods and Performance Measurement
12.1
Forecasting Methods
12.1.1
Using Averages
12.1.1.1
Simple Averages
12.1.1.2
Moving Average (MA)
12.1.1.3
Double Moving Average
12.1.2
Linear regressions
12.1.3
Non linear regressions
12.1.4
Smoothing methods
12.1.4.1
Exponential smoothing
12.1.4.2
Holt’s exponential smoothing
12.1.4.3
Winters’ exponential smoothing
12.1.4.4
Moving Averages, see section @ref(MA)
12.1.5
ARIMA
12.1.6
Dynamic forecasting
12.1.6.1
ADL
12.1.6.2
VAR
12.1.7
Decompisition
12.2
Performance Measurements
12.2.1
Error terms
12.2.2
Multicollinearity
12.3
Statistical tests
12.4
Formulas
12.5
Loops!!!!
12.5.1
HoltWinters Smoothing - finding optimal frequency
12.5.2
ARIMA - it is not really that useful, does not have all combinations
Business Forecasting
12
Methods and Performance Measurement