Improved Forecast Errors for TCF Commodity Outputs: version
Figure 6: 1992 ‐ 1998 – U.S trade by the Knitfabric industry in nominal dollars
1. Why did the model erroneously give good prospects to Apparel ?
Apparel had a USAGE error of 121% versus the larger trend forecast error of 148%. The key results for this commodity are shown in Table 17. The actual outcome for Apparel output (x0dom) was a 47.6% contraction over the 1998‐2005 period. This followed 25.0% growth from 1992‐1998. The extrapolated trend was therefore 30% growth versus the USAGE forecast of 15.8% growth. Table 19 shows the main users, cost structure and other information of interest of the 1998 database that was used in the forecast. The following observations can be made:
About 49% of total domestic sales came from imports (Section 5 of Table 19).
Just 11% of domestic production was exported (Section 3 of Table 19).
87% of domestic output was sold to consumers (Section 5 of Table 19).
A priori, given the strong share of sales to consumers, an inaccurate projection for the household taste variable (a3com) could create a material divergence between forecast and reality. The model extrapolated household preferences from the historical run resulting in tastes moving in favour of Apparel. This did not turn out to be true, with a difference of 24.6 percentage points. However, no clear evidence could be found to suggest that by 1998 consumer tastes were souring towards this commodity.
More important was the model’s underestimation of impftwist. As mentioned previously, this purports to measure the impact of the shifter on the twist (ftwist_src). This can be seen in Table 18. The impact on the domestic sales share of Apparel was a 12.3% contraction in forecast, ceteris
paribus, when in reality the damage to the market share of domestic producers was 57.8%.
ftwist_src 29.87% ftwist_src 291.82%
start end start end
Import 3211 21.44 26.17 Import 3211 21.44 51.68
Domestic 11764 78.56 73.83 ‐6.02 Domestic 11764 78.56 48.32 ‐38.49
Total 14975 = BAS1 Total 14975 = BAS1
start end start end
Import 0 0 0 Import 0 0 0
Domestic 0 0 0 0.00 Domestic 0 0 0 0.00
Total 0 = BAS2 Total 0 = BAS2
start end start end
Import 51931 53.38 59.79 Import 51931 53.38 81.77
Domestic 45350 46.62 40.21 ‐13.75 Domestic 45350 46.62 18.23 ‐60.90
Total 97281 = BAS3 Total 97281 = BAS3
Weighted Average = ‐12.72 Weighted Average = ‐57.91 Sales 112256 impftwist (%) ‐12.33 Sales 112256 impftwist (%) ‐57.83
Apparel (original forecast) Apparel (actual outcome)
The difference is due to b.o.t.e. estimation error The difference is due to b.o.t.e. estimation error Table 18: The predicted and actual impacts of import twist factors on Apparel in the period from 1998 to 2005 In addition, reduced protection coincided with the increasing emergence of China and India as super‐cheap producers and exporters of Apparel. This is reflected in the trade data, which is illustrated in Figure 7. As was the case for all TCF commodities, the foreign currency price increase of imported Apparel that occurred between 1992 and 1998 was projected forward. This strongly contributed to a higher basic price of imported Apparel than turned out to be the case (in fact, the
basic import price fell sharply). This impacted relative basic prices in a way that was incorrectly favourable to domestic producers. The model estimated an 11.7% change in favour of domestic output, when in reality there was an unfavourable 13.0% change.
Finally, the impact on output of the sharp cost reductions relating to primary factor input that occurred from 1992 to 1998 were also projected forward; but instead of these intensifying they failed to materialise. 0 5 10 15 20 25 30 0 10000 20000 30000 40000 50000 60000 1992 1993 1994 1995 1996 1997 1998
%
$m
Trade Data ‐Apparel & Other Textile Products
Value of imports Value of exports %chg Imports %chg Exports
Figure 7: 1992‐1998 – U.S. trade by the Apparel & Other Textile Products industries in nominal dollars
Table 19: The key attributes of Apparel in 1998
2. Macro perspective
The impact of the Uruguay Round of multilateral trade negotiations (MTN) that was implemented over the period 1995‐2000 for developed countries has already been discussed. As part of this the
U.S. lowered tariffs and commenced the phaseout of quotas but reserved the right to impose safeguards once the phaseout was complete. However, the U.S. refused to agree to accelerated quota growth and tariff reductions. Furthermore under the terms of the Uruguay Round agreement, developing countries were afforded much higher tariff rates than developed countries. The modeller could not have been sure that reductions in protection would subsequently be reversed. In forming a view about the prospects for Apparel the modeller could have noted the advice in the following quote:
“Key factors that affect the demand for many textiles and apparel subsectors include health of end‐use markets, growth in the overall economy and consumer spending, and trends in foreign trade. A broad range of textile products are used in the production of apparel, home furnishings, and industrial products ... Most apparel markets are affected
by trends in consumer spending, overall growth in the national economy,
demographics, and foreign trade.”46
By late 1998 it was not clear that consumer tastes would begin to sour overall, whilst taking an increased liking to imports – well beyond that which could be explained by changes in relative prices.
3. Conclusion
In summary, import twist factors, relative prices and household preferences all worked against domestic output of Apparel, resulting in a large forecast error. Whilst the modeller could have been wary of strong domestic growth numbers given that TCF industries were becoming increasingly exposed to foreign entry, it seems that it would have been too difficult to predict the magnitude of the preference twist in favour of imports. By late 1998 it was not clear that consumer tastes would begin to sour overall, whilst taking an increased liking to imports – well beyond that which could be explained by changes in relative prices. Furthermore, the modeller could not have been sure that reductions in protection would not subsequently be reversed. However, as was the case for all TCF commodities, improvements could have been made to the import price forecasts because basic import prices were heavily tied to policy.
4. Strategy to improve the forecast
In re‐running the simulation, real basic import prices were projected forward, generating more realistic relative basic price changes. For Apparel this produced an improved forecast for output growth of 3.8% and reduced the USAGE error from 121% to 98%. This was done for all TCF industries, which resulted in more realistic import price projections and hence, less erroneous domestic‐import relative price movements. This improved the estimate for x0dom_dom and, in
turn, for x0dom. However, the absolute size of the error remained quite large due to the ongoing underestimation of impftwist, the error in forecasting household preferences, and the overestimation of factor input cost reductions. A subsequent simulation differing only by the additional forecast of no further primary factor cost savings (i.e., no further all factor augmenting technical change) gave better results. (See earlier comments for elaboration.) In this case, higher costs meant that output contracted 24.7% and the USAGE error improved to 44%.
Luggage → Luggage (SIC 3161)
Establishments primarily engaged in manufacturing luggage of leather or other materials. The
luggage industry produces a wide variety of products, including suitcases, briefcases, attaché cases,
hand luggage, tote bags, trunks, and occupational cases. Materials used in addition to leather
include plastics, nylon, cotton, linen, and metals. Many products use a combination of these
materials. Construction methods include sewing, molding, and laminating.
Table 20: Key results for Luggage