FORTRAN subroutines for time series analysis

; The authors have recently been concerned in a time-series study that constituted a fairly typical piece of applied statistical i ~J research, involving extensive computations on a moderately large quantity of data. We have found that the many different numerical processes that were required could be built up almost completely from a small number of basic operations, and a set of FORTRAN subroutines has been written to perform these. The main purpose of this note is to describe these sub-routines, but since the question of general statistical programs is topical [1 ], we include some general remarks. The desirability of a "building-block" type of approach, made possible by the FowrRax subroutine facility, is a :i matter for argument. This approach amounts to the use of an interpretive programming technique, as opposed to an ::i assembler or compiler. A subroutine call is analogous to a ;: high-level instruction which is decoded into machine : language each time it is encountered in a program. The chief advantage of interpreters lies in the fact. that they are far simpler to write than compilers. This is over-whehningly so in the present application in which the vast bulk of the programming work was taken care of by I.~' ORTRAN. It is also a matter of great practicM importance; the set of subroutines here described evolved as a by-product of an actual statistical investigation in which the time available for programming was severely limited. In the construction of the subroutines and in their subsequent use, we have gotten through a very large amount of computation without imposing a load on scarce programming resources. Interpreters are usually held to be inefficient because of the individual handling of each high-level instruction and the unproductive time spent in interpretation. The importance of these considerations depends to a large extent upon the size of the building blocks. F(mTaAN has a eertMn capability for tackling a programming p r o b l e m as a whole rather than piecemeal , and this can be g i v e n full play within each subroutine. Equally, the i m p o r t a n c e of interpretation time depends upon the proportion which this bears to computing time. Both these considerations make it desirable for each subroutine to carry out. a fairly substantiM piece of computation. The opposite l i m i t a t i o n …