2.1 Proyección térmica
2.1.6 Recubrimientos
2.1.6.3 Recubrimiento Cu-Al (Proxon 21071)
Following the data selection, cleansing process and application of IBES internal rules, initial results of descriptive results of analyst forecast accuracy revealed extremely high values of cross-sectional analysts’ actual forecast error96 (AEAF_ACC) throughout the 11 months
prior to reported earnings. For example, the actual forecast errors (AEAF_ACC_ACTUAL) shown in Appendix A1 were 126.88% at 11 months prior to reported earnings and were 72.44% at 1 month prior to reported earnings. This prompted an investigation into the possible causes of such high actual forecast error. The aim was to determine appropriate ex- ante data partitioning criteria that could be applied to the initial 14 years data set (1st July
1988 through 30th June 2002) and in turn reduce the per month prior reported earnings actual
forecast error.
The 14 year data set was first partitioned into two time periods. The first time period runs from 1st July 1988 through 28th February 1999 and the second time period, which represents the onset of the bursting of the internet bubble, runs from the 1st March 1999 through 30th June 200297. It was expected that the forecast errors of the economic cycle downturn
induced by the bursting of the internet sector bubble would be significantly higher. The unexpected downturn in the technology sector produces more earnings forecast surprises, which in turn contributes to a substantially higher AFE.
Another possible cause of the high AFE was the high forecast error associated with lower market capitalised firms. It was expected that lower market capitalised firms would have lower levels of analyst coverage and also lower volume of information content disseminated. These factors would result in inaccurate (high error) forecasts being made.
The final factor that may contribute to the high AFE in the initial results may be a result of high forecast errors associated with specific industries such as gold mining and industries with substantially low analyst coverage. Table 5-5 shows the change in sample size as each
96The actual forecast error (AFE) is defined as the absolute difference between the actual announced earnings and forecast divided by the actual announced earnings.
97These economic cycle periods can be obtained from the Economic Cycle Research Institute (ECRI).
of the partitioning criteria just described is applied to the 14 years (1988-2002) data set (see Table 5-1 for sample size changes from the original data set to the 14 years data set).
Table 5-5. Sample Size Changes Due to Subsequent Application of Selection Criteria
Criteria Maximum Number of Unique Firms Number of IBES Individual Analyst’s Annual Earnings
Forecasts Number of IBES Consensus
(1) Select data from years
1988 through 2002 3,939 268,561 39,028 (2) Removal of 1st March, 1999 to 30th June 2002 forecasts 2,431 207,403 27,382 (3) Removal of bottom 50% market capitalised firms 1,732 185,341 20,496
(4) Remove gold mining
industry 1,507 161,168 17,977 (5) Removal of bottom 75% market capitalised firms 907 123,120 10,843
A number of patterns emerged in the IBES consensus actual forecast error results from the application of the time period, firm size and industry type partitioning criteria just described. For the time period AFE analyses, the 14 year data set is partitioned into four consecutive economic cycle time periods obtained from the ECRI: 1st July 1988-30th September 1994, 1st
October 1994-30th September 1996, 1st October 1996-28th February 1999 and 1st March 1999-30th June 2002. The fourth period represents the economic downturn induced by the
internet dot com bust period. Appendix A2 presents the AFE for the four periods at 1 month forecast horizon and the average over all 11 months for both profit (positive actual reported EPS) and loss (negative actual reported EPS) firms. For profit firms, IBES consensus mean AFE is the highest for the dot com bust period at 1 month forecast horizon (47.39%) compared to other periods. For loss firms, IBES consensus mean AFE is also the highest for the dot com bust period at 1 month forecast horizon (242.54%) compared to other periods. These results are consistent with the expectation that the internet downturn contributes a relatively higher AFE to the overall 14 year period AFE.
Appendix A3 presents IBES Consensus AFE by market capitalisation. In general, AFE decreased with increasing firm size. At one month forecast horizon, first quartile (lowest 25% market capitalised firms) IBES consensus mean actual forecast error is 85.99%. This
reduces to 51.10%, 51.27% and 19.67% for the second, third and fourth quartile respectively. Thus, as expected, smaller firm size is associated with a higher AFE. The high forecast error difference between the first and second quartiles (85.99%-51.10%=34.89%) leads to the decision to remove the bottom 50% market capitalised firms as shown by row (3) in Table 5-3. As a result 185,341 individual analysts’ earnings forecasts and 20,496 IBES consensus forecasts remain.
A sector/ industry AFE decomposition between profit and loss firms partitioned by a breakdown of the 14 year data set into the three consecutive ECRI economic cycles and the dot com bust period is shown in Appendix A4. Highlighted in this table are the AFEs of the Gold Mining industry. The negative actual EPS firms of the 1st July 1988-30th September 1994 economic cycle within the Gold Mining industry for the third quartile market capitalised firms has an AFE of 108.39%. This figure increased further for the consecutive period 1st October 1994-30th September 1996 (134.93%) and the 1st October 1996-28th
February 1999 period (85.61% at third quartile and 384.21% at fourth quartile market capitalisation). During the bursting of the internet bubble (1st March 1999-30th June 2002),
the same sub-data set has AFEs of 126.35% and 83.33% for the top two market capitalisation quartiles respectively. These results confirm that large Gold Mining firms significantly contribute to the overall AFE. Thus gold mining firms are removed from the sample described by row (3) in Table 5-3 with the resulting sample reflected by (4). This sample contains 185,341 individual analysts’ earnings forecasts and 20,496 IBES consensus forecasts.
Finally, a decision to remove the bottom 75% of market capitalised firms rather than the removal of bottom 50% market capitalised firms was made to further reduce the actual forecast error in the sample. The application of all the criteria combined yields 907 unique firms per forecast horizon, 123,120 monthly earnings estimates and 10,843 monthly IBES consensus figures.
Therefore, the partition analysis has shown that small firm size, gold mining industries and the internet bust business cycle adds significantly high AFE to the overall AFE of the 14 year data set. To that end, the data are culled using these three criteria and it will be shown in Chapter 6 results that monthly analysts’ actual earnings forecast error does indeed decrease significantly. Due to the culling of the 14 year sample data, a rerun of all
hypothesis tests within the three Phases of this thesis will be carried out using the culled data set. The next section concludes the chapter.