| 年份 | 多元回归预测误差值 | 神经网络预测误差值 |
| 2000 | −2311.91 | −4.322 |
| 2001 | −2022.79 | −4.000 |
| 2002 | −1740.59 | −3.678 |
| 2003 | −1052.8 | −3.356 |
| 2004 | −158.691 | −3.304 |
| 2005 | 293.064 | −2.713 |
| 2006 | −1539.66 | −2.391 |
| 2007 | 2576.228 | −2.069 |
| 2008 | 3696.241 | −1.747 |
| 2009 | −4187.03 | −1.425 |
| 2010 | −1535.67 | −1.103 |
| 2011 | 610.2513 | −0.7813 |
| 2012 | 1925.462 | −0.4595 |
| 2013 | 3164.174 | −0.1376 |
| 2014 | 965.4246 | 0.1842 |
| 2015 | −1456.15 | 0.5061 |
| 2016 | −3794.2 | 0.828 |
| 2017 | −3341.61 | 1.15 |
| 2018 | −2124.55 | 1.472 |
| 2019 | −1735.91 | 1.794 |