|
<< Click to Display Table of Contents >> ITERSHOW |
  
|
|
<< Click to Display Table of Contents >> ITERSHOW |
|
The statement ITERSHOW show details of previous Gauss-Seidel iterations (cf. SIM).
itershow < period > variables;
period |
(Optional). Local period, for instance 2010 2020, 2010q1 2020q4 or %per1 %per2+1. |
variables |
Variable(s) or a list like {#m}. |
Use this syntax to show iterations for the variable gdp for the year 2010:
itershow <2010 2010> gdp; |
The statement produces output containing (for each iteration) the value of the endogenous variable before and after simulating the Gauss-Seidel loop, and differences, different criteria etc.
The output shows the iteration number, values before and after the iteration. The next column "Hist. var" is the historical variance/variability in the data, obtained by means of looking at lagged historical values of the endogenous variables.
The next column is the difference between the values before and after the iteration, and "Relative1" is this difference divided by the historical variance. This is the criterion Gekko uses for relative convergence per default.
The last column is the difference divided by the value before the iteration. "Relative2" is the criterion used in the software package PCIM. So "Relative1" corresponds to setting "OPTION solve gauss conv = conv1" (default), and "Relative2" corresponds to setting "OPTION solve gauss conv = conv2".
It is often the case that "Relative1" is larger than "Relative2", so using this criterion is stricter and would demand more iterations given the same relative criterion (0.0001 for instance). Sometimes the inverse is true, however, especially when the "Before" value is close to zero, whereas the historical variability is large. In that case, the PCIM-like criterion ("Relative2") would be stricter. But often this will just postpone the solution, in case the variable just happens to be have a solution close to zero in that particular year (but without being close to zero in general). Examples of this could be balances and flows, for instance the balance of payments, net investments, revaluation ("omvurderinger") etc. Close-to-zero solutions for such variables are different in kind from variables with levels generally close to zero (in ADAM, for example interest rates). Looking at historical variability as done in Gekko is actually a means to try distinguishing such classes of variables from each other regarding convergence. If, for instance, the balance of payments can change by an amount of around 20000 (million DKK) from year to year, and the true solution in a particular year just happens to be 1 million DKK, we are not interested in obtaining an extreme precision (i.e., many digits after the decimal point) regarding that particular value. If the variable normally can change by an amount of 20000 from year to year, a solution of 2 is nothing to worry about, even if the true solution is 1. Whereas if the true solution regarding an interest rate is 0.04, an alternative solution of 0.08 is worrying.
In order for this statement to work, option solve gauss dump must first be set to yes, and a simulation performed. Beware that setting this option consumes a lot of RAM when simulating, and also slows simulations down. In case of a RAM error, try to limit narrow the time period.