The input coefficient matrix is a very important variable in input-outputanalysis. You must use estimates when you want to use it because published input-output tables always lag by about two or three years. Most of the matrix methods estimate need data from the National Income Account. But the National Income Account can’t offer all the datas. That is, you must estimate part of the data which is used in the matrix estimate method.
This paper obtains the time series data of elements of the input coefficient matrix from the published input-output tables. Regression of each series is carried out to twenty functional forms in the variable of time trend. If we set the value of time trend in the best fitting function of each series, then we can obtain the projected input coefficient matrix.
As to the performance of this approach. I tested by the statistics of Absolute Average Percentage Error (AAPE), Root Mean Square Error (RMSE), and Root Mean Square Percentage Error (RMSPE). I found that AAPE improves significantly, RMSE improves just a little, but RMSPE is reduced by more than half.
