MODGAMMA

The MODGAMMA option specifies a value to use in testing how much an equation in a simultaneous block of a model is changing between iterations. MODGAMMA controls the degree to which the test compares the absolute amount of the change between iterations versus the proportional change. MODGAMMA is especially important in testing equations that result in very small values.

Data type

INTEGER

Syntax

MODGAMMA = {n|1}

Arguments

n

An integer value to use in testing for convergence and divergence. As Oracle OLAP calculates each equation in a simultaneous block, it constructs a comparison value that is based on the results of the equation for the current iteration and the previous iteration. When the comparison value passes a tolerance test, the equation is considered to have converged. When the comparison value meets an overflow test, the equation is considered to have diverged.

The comparison value that is tested is as follows.

(thisResult - prevResult) DIVIDED BY (prevResult PLUS MODGAMMA)

where thisResult is the result of this iteration and prevResult is the result of the previous iteration.

Oracle OLAP calculates the absolute value of the enclosed expression. The default value of MODGAMMA is 1.

Notes

In the test for convergence, the MODTOLERANCE option determines how closely the results of an equation must match between successive iterations. With the default value of 3 for MODTOLERANCE, the equation is considered to have converged when the comparison value is less than 0.001.

In the test for divergence, the MODOVERFLOW option determines how dissimilar the results of an equation must be in successive iterations. With the default value of 3 for MODOVERFLOW, the equation is considered to have diverged when the comparison value is greater than 1000.

The comparison value that Oracle OLAP evaluates in tests of convergence and divergence is fundamentally a proportional value. It expresses the change between iterations as a proportion of the previous results. In the test for convergence, the change between iterations must be small in proportion to the previous results. In the test for divergence, the change between iterations must be large in proportion to the previous results. By testing a proportional value, rather than testing the absolute amount of change, Oracle OLAP can apply the same test criteria to all equations, regardless of the magnitude of the equation results.

However, the comparison value that Oracle OLAP tests is not strictly proportional. When the results of an equation are very close to zero, the denominator of a strictly proportional comparison value would also be very close to zero, and thus the comparison value itself would generally be large. Therefore, the test for convergence would be difficult to satisfy, while the test for divergence would be easy to meet. To solve this problem, Oracle OLAP adds the value of MODGAMMA to the denominator of the comparison value. When the default value of 1 is used for MODGAMMA, the effect of MODGAMMA is as follows:

• When the equation results are close to zero, the denominator is close to one and the test is essentially a test of the absolute change between iterations.

• When the equation results are very large, the effect of adding MODGAMMA to the denominator is negligible, and the test is close to being a strictly proportional test.

For equation values close to zero, you can control the sensitivity to the tests for convergence and divergence by changing the value of MODGAMMA. When equation values are very small, you essentially scale the changes in model values between iterations when you change the value of MODGAMMA. For example, when you change MODGAMMA from 1 to 2, the comparison value is essentially cut in half. As a consequence, you reduce the likelihood that divergence will occur.

The default value of MODGAMMA is appropriate in most situations. When you increase the value of MODGAMMA, the model equations will converge more quickly, but the results will be less precise. The smaller the equation value, the more pronounced is the effect of increasing MODGAMMA; convergence is attained relatively more quickly for small model values, while more precision is lost.

You can also force the simultaneous blocks of a model to converge more quickly by decreasing the value of MODTOLERANCE and thereby relaxing the test for convergence. However, when you do this, you sacrifice the precision of all the results, not just the results of equations with small values.

Therefore, when a model contains some equations with large values and some equations with very small values, it might be preferable to increase MODGAMMA rather than decreasing MODTOLERANCE. By increasing MODGAMMA, you might be able to force equations with small values to converge more quickly while retaining the precision of equations with large values.

Examples

MODTRACE = YES
MODSIMULTYPE = 'GAUSS'
LIMIT division TO 'Camping'
income.bud budget

(MOD= INCOME.BUD) BLOCK 1: SIMULTANEOUS 
   ...
(MOD= INCOME.BUD) BUDGET (LINE NET.INCOME MONTH 'JAN97' ITER 16) = 0.026243533
     ...
(MOD= INCOME.BUD) BUDGET (LINE NET.INCOME MONTH 'JAN97' ITER 17) = 0.024054312
   ...
(MOD= INCOME.BUD) BUDGET (LINE NET.INCOME MONTH 'JAN97' ITER 18) = 0.025788293
   ...
(MOD= INCOME.BUD) BUDGET (LINE NET.INCOME MONTH 'JAN97' ITER 19) = 0.024390642
    ...
(MOD= INCOME.BUD) BUDGET (LINE NET.INCOME MONTH 'JAN97' ITER 20) = 0.025501664
    ...
(MOD= INCOME.BUD) BUDGET (LINE NET.INCOME MONTH 'JAN97' ITER 21) = 0.024608562

(0.024608562967 - 0.025501664970 = 0.00087) / (0.025501664970 + 1)

MODGAMMA = 2
income.bud budget

(0.025788293304 - 0.024054312748 = 0.00086) / (0.024054312748 + 2)