1. Recap: How the Process Works (1) Determine the weights. The weights can be absolute or relative. Weights encompass two parts -- the quantitative weight and the current evaluation of its importance towards explaining the objective. (2) Once the weighting scheme is determined [shown in matrix A], solve the linear equation (or use approximation methods): A w = max w that is, (A - max I) w = 0 If this equation has a nonzero solution for w, then max [which is a scalar] is said to be an eigenvalue or characteristic value of A [which is an n x n matrix of pairwise comparisons] and w [which is an n x 1 matrix] is said to be an eigenvector belonging to. I is the identity matrix, which is a diagonal matrix with the main diagonal terms equal to 1 and zero elsewhere. (3) The solution provides the answer to the most likely outcome, given your judgmental rankings of all the individual criteria.

2. The Problem Setup: Form Matrix of Ratio Comparisons and Multiply by w

3. Approximation & Exact Methods to Derive a Solution 1. Normalize the geometric means of the rows. This result coincides with the eigenvector solution when n [ 3. 2. Normalize the elements (first) in each column of the judgment matrix and then average over each row. A simple way to obtain an estimate of max, if the exact value of w [an n x 1 matrix] is available in normalized form, is to add the columns of A and multiply the resulting vector by the priority vector w. 3. Use linear algebra to solve for the eigenvector and normalize the values for max.

4. Cell J9 is the geometric average ---- =(A9*B9*C9*D9*E9*F9*G9*H9)^(1/8) A Short Cut for Determining Weights Comparison of the Geometric Average Approximation and Exact Linear Algebra Calculation for an 8 x 8 Matrix

5. Another Method: Solving for the Weights by Successive Squaring and Checking Differences

6. Second Iteration

7. Compare the Results Using a Statistical Package (RATS) to Calculate the Eigenvector.

8. How consistent are the inputs? In the application of AHP it is possible that inconsistent judgments of input desirability may have crept into the process. For example, the original matrix, A, may not have full transitivity -- that is, A(1) may be preferred to A(2) and A(2) to A(3), but A(3) may be preferred to A(1). To determine how much inconsistency is in the A matrix, Saaty defined a measure of deviation from consistency, called a consistency index, as: C.I. = ( max - N)/(N-1), where N is the dimension of the matrix. Then, Saaty calculated a consistency ratio (C.R.) as the ratio of the C.I. to a random index (R.I.) which is the average C.I. of sets of judgments (from a 1 to 9 scale) for randomly generated reciprocal matrices.

9. Saaty’s Calculated Random Index Measures for Various Sizes of “N” Average Random Consistency Index (R.I.) n 1 2 3 4 5 6 7 8 9 10 R.I. 0 0.52.89 1.11 1.25 1.35 1.40 1.45 1.49 --------------------------------------------------

10. AHP Diagnostics

11. Example 1: Real GDP Growth Forecasting From the Perspective of Late 1991 Based on Saaty and Vargas, “Incorporating Expert Judgment in Economic Forecasts: The Case of the U.S. Economy in 1992”

12. Definitions Used Conventional adjustment assumes a status quo with regard to the system of causes and consequences in the economy. Economic restructuring assumes a new environment.

13. Adding Another Level of Subfactors to the AHP Model

14. The Expanded AHP Model Once the Structure of the Model is Formulated, the Judgmental Weights Must be Assigned for a Given Forecast Horizon...

15. Which primary factor will be most influential in determining the strength of the recovery? Conventional Adjustment (CA)Restructuring (R) Conventional Adjustment (CA)11/5 Restructuring (R)51 Vector Weights 0.167 0.833 Step 1: Determining Weights Top Down Determining the First Weight Using the Geometric Average Step 1: 1 x (1/5) = 0.20 Step 2: 0.20 ^ (0.5) = 0.447 Step 3: Normalize Weights (0.447/2.683) =0.167 Note: In order to Normalize the Weight in Step 3, you need to know that the second row geometric average is 2.236.

16. Which subfactor is more important in influencing conventional adjustment? Step 2-A: Determining Weights Top Down

17. Step 2-B: Determining Weights Top Down Which subfactor is more important in influencing restructuring? Follow the Same Process for Next Level...

18. Vector Weights for Other Alternatives Based on Saaty’s Example Relative likelihood of the strength of the recovery if consumption drives the expansion under conventional adjustment Relative likelihood of the strength of the recovery if investment drives the expansion under conventional adjustment Relative likelihood of the strength of the recovery if exports drives the expansion under conventional adjustment Relative likelihood of the strength of the recovery if confidence drives the expansion under conventional adjustment VS = 0.423 S = 0.423 M = 0.104 W = 0.051 VS = Very Strong Growth S = Strong Growth M = Moderate Growth W = Weak Growth VS = 0.095 S = 0.095 M = 0.249 W = 0.560 VS = 0.182 S = 0.182 M = 0.545 W = 0.091 VS = 0.376 S = 0.376 M = 0.193 W = 0.054

19. Vector Weights for Other Alternatives Based on Saaty’s Example Relative likelihood of the strength of the recovery if fiscal policy drives the expansion under conventional adjustment Relative likelihood of the strength of the recovery if monetary policy drives the expansion under conventional adjustment VS = 0.125 S = 0.125 M = 0.625 W = 0.125 VS = Very Strong Growth S = Strong Growth M = Moderate Growth W = Weak Growth VS = 0.084 S = 0.084 M = 0.649 W = 0.183 Calculate the Weights for the Economic Restructuring Alternative...

20. Vector Weights for Other Alternatives Based on Saaty’s Example Relative likelihood of the strength of the recovery if financial sector drives the expansion under restructuring VS = 0.095 S = 0.095 M = 0.249 W = 0.560 VS = Very Strong Growth S = Strong Growth M = Moderate Growth W = Weak Growth VS = 0.055 S = 0.118 M = 0.262 W = 0.565 Relative likelihood of the strength of the recovery if defense posture drives the expansion under restructuring Relative likelihood of the strength of the recovery if global competition drives the expansion under restructuring VS = 0.101 S = 0.101 M = 0.348 W = 0.449

21. GDP Forecasting Model Once the Structure of the Model is Formulated and the Judgmental Weights Assigned, the Forecast Now Can Be Determined...

22. Overall Results of the Forecast Based on Mid-Points of Forecast Ranges Using Matrix Multiplication Obtain the Forecast as Follows:

23. This Same Model Could Have “Alternatives” as a Time Horizon Instead of Strength of Recovery Three Months Six Months 12 Months 24 Months or More Model Could Be Used for Various Forecast Horizons

24. Example 2: Forecasting Foreign Exchange Rates Six Primary Indicators of Future Spot Exchange Rates: (1) Relative interest rates (INTRAT); (2) Forward exchange rate biases (FDBIAS); (3) Official exchange market intervention (EXCINT); (4) Relative degree of confidence in the U.S. economy (CONFUS); (5) the size and recent direction of the U.S. current account balance (CURBAL); (6) Past behavior of exchange rates (PASREC). Source: Saaty and Vargas, Prediction, Projection and Forecasting, Kluwer Academic Publishers, Norwell, MA, 1991.

25. Yen/Dollar Exchange Rate Forecasting Hierarchy GOALForex Value of the Yen in 90-Days Three Levels of Criteria Level 1: (A) Relative interest rates (INTRAT); (B) Forward exchange rate biases (FDBIAS); (C) Official exchange market intervention (EXCINT); (D) Relative degree of confidence in the U.S. economy (CONFUS); (E) the size and recent direction of the U.S. current account balance (CURBAL); (F) Past behavior of exchange rates (PASREC) Level 2: (A) Fed Policy; (B) Size of Federal Deficit; (C) BoJ Policy; (D) Consistent/Erratic Forex Intervention; (E) Forward Rate Premium/Discount; (F) Size of Forward Rate Differential; (G) Relative Interest Rates; (H) Relative Growth; (I) Relative Political Stability; (J) Size of Current Account; (K) Expected Change in Current Account; (L) Past Exchange Rate Relative/Irrelative Level 3: Generally of the form: (A) Higher/Tighter, (B) Neutral, and (C) Lower/Easier Sharp Decline / Moderate Decline / No Change / Moderate Increase / Sharp IncreaseForecast

27. Some Other Applications of AHP to the Forecasting Problem To Address the Duration or Time to Event Question. To Assess the Likelihood of Strength or Weakness. To Determine Judgmental Probabilities. To Judge the Likelihood of a Cyclical Turning Point. Could Be Used to Develop a Consensus Forecast Based on Alternative Techniques or Inputs.