Hello Everyone,
I have to calculate the predicted values for year 2015, 2016 and 2017 for Bucket1, Bucket2, Bucket3 and Bucket4 using a Z_score and an Index value. I already have Z_Score from other table, but I need to generate an Index value (which is a random value).
The Index value to be generated should be such that that when we calculate predicted PD using it for each year. The sum square errors of Predicted PDs and Actual PDs should be minimum.
I am able to do this in Excel using Run solver function, and I need to do this using PROC FCMP in BASE SAS in SAS EG as I don't have SAS/IML or SAS/OR COMPONENT. Please help me with this code as I am new to PROC FCMP.
Attached is the Input table and desired output table:
Table 1: Using Z Score from this
| Col1 | Value 1 | Z value |
| 1 | 0.86% | -2.38 |
| 2 | 26.02% | -0.64 |
| 3 | 42.43% | -0.19 |
| 4 | 64.78% | 0.38 |
Table 2: Bucket-wise Actual values for each year
| Col1 | 2015 | 2016 | 2017 | 2018 |
| 1 | 1.15% | 0.66% | 0.51% | 1.13% |
| 2 | 32.38% | 22.38% | 18.04% | 31.26% |
| 3 | 50.08% | 37.32% | 38.41% | 43.91% |
| 4 | 69.31% | 59.29% | 57.13% | 73.38% |
| 5 | 99.72% | 100.00% | 100.00% | 100.00% |
Table 3: Index values are random values generated, and used in calculating predicted values in bold
| Index value | 0.100000 | 0.170699 | -0.133416 | -0.178382 | 0.135047 |
| Col1 | | 2015 | 2016 | 2017 | 2018 |
| 1 | | 1.35% | 0.59% | 0.52% | 1.23% |
| 2 | | 31.84% | 21.88% | 20.58% | 30.58% |
| 3 | | 49.19% | 37.28% | 35.59% | 47.77% |
| 4 | | 70.88% | 59.71% | 57.96% | 69.65% |
Table 4: Min sum of squares (between actual values and predicted values) value and Index value.
| | Sum of Squares | Index |
| | 0.000 | |
| 2015 | 0.000 | 0.17 |
| 2016 | 0.000 | -0.13 |
| 2017 | 0.002 | -0.18 |
| 2018 | 0.003 | 0.14 |
I have written the below code using proc fcmp for the above problem, but I am not able to create the right code for this:
PROC FCMP OUTLIB=work.functions.ci;
function predict(currentValue);
return(CDF('NORMAL', currentValue + index));
endsub;
function sum_square_error(value1,value2);
diff = (value1-value2);
return(diff*diff);
endsub;
function minimum(y,sse,index,value1,value2);
return(solve("sum_square_error",.,y,min(sse));
endsub;
QUIT;
options cmplib=( work.functions);
data test;
set test;
predicted = cred_index(Z_score);
err = sum_square_error(actual,predicted);
run;
Please help with this problem.
