This paper deals with the size of the random walk property of Colombia's output in two periods 1925-1994 and 1950-1994. GDP and GDPPC were both found to be integrated of order one a result which is very well known. The sequences are highly persistent, especially in the period 1950-1994. The forecast error when an innovation of 1 percent enters into the economy is about 1.5 percent in the very long run, when GDP is considered. The response is about 1.3 percent in the case of GDPPC, which seems to give support to the idea that population growth is a source of nonstationary in some macroeconomic aggregates. For the larger sample (1925-1994) persistence is less. This result could cast some doubt on the method of estimation of GDP for the period 1925-1950. Finally, evidence of nonlinearity is found only in Hodrick- Prescott filtered variables dated between 1925 and 1994. This leaves open the question about whether the HP filter introduces nonlinearity in the high frequency variable that it generates.
