| The purpose of this study was to assess the ability of empirical models derived from the translog (TL) approximation to track the characteristics of known underlying technologies with increasing complexity, including technological progress.; In the investigation, not only the tracking ability of the flexible models was measured but also factors affecting their performance were identified, as well as methods for improvements.; Knowledge of the data generating process permits better insight into the approximating ability of a flexible form than the empirical methods using actual data.; Applying Monte Carlo methods, models using data with and without errors were estimated in order to distinguish between problems attributable to the tracking ability of the flexible forms apart from biases caused by errors-in-variables. Furthermore, the effects of different hypotheses, restrictions, flexibilities and estimation procedures on the ability of the TL functional forms to capture characteristics of the true technology were studied.; When the data were measured without errors in variables, the TL models performed, in general, very well in terms of bias and mean absolute deviation (MAD) of input substitution elasticities, returns to scale, rates of technical change and total factor productivity growth. Furthermore, all the models tracked accurately biased technical change and satisfied regularity conditions.; Disparities of estimates obtained from several models and substantial increases in biases and MAD errors, over those reported under a regime without measurement errors indicated performance deterioration caused by mismeasured data.; Performance gains by using the instrumental variable estimation method (I3SLS), even with large errors-in-variables, indicated that simultaneity bias caused by mismeasured data is a problem in addition to errors in data, and it should not be ignored. Furthermore, failure of flexible models to satisfy regularity conditions may be due to measurement errors and to inappropriate estimation method.; The superiority of the transformed TL cost models under different error regimes and estimation methods indicates the usefulness of transforming the data to deviation from means.; The results suggest that improvements in tracking performances of flexible models are likely to be achieved through better quality of data and improved methods of estimation dealing with errors. |
