美国失业率计算有误，实际情况更令人担忧
美国总体失业率从5月的13.3%下降到6月的11.1%。这是好消息。今年4月，美国失业率达到14.7%，创自二战以来的新高。失业率连续下降意味着遭到公共卫生事件严重冲击的美国就业市场，似乎正在快速复苏。
此外，美国失业率好转的情况，比总体失业率所表现出来的趋势更令人高兴，这似乎也是一条好消息。然而不幸的是，失业率好转的原因是，上个月进行失业率对比的起点要比13.3%高很多。
美国劳工统计局（U.S. Bureau of Labor Statistics）在7月2日发布的新闻稿的脚注中写道，总体失业率可能低估了真实失业率，因为数据分类错误导致其每月家庭普查数据将临时失业者划分为就业者。劳工统计局在4月3日、5月8日和6月5日分别针对3月、4月和5月的就业情况发布的新闻稿中，也加了类似的脚注。
以下是对这项误差的解释。美国劳工统计局在每个月中旬进行家庭普查，利用普查结果计算出每个月的失业率，每个月结束后会很快对外公布。劳工统计局在脚注中解释说，其曾经指示普查人员将因为与疫情相关的企业临时停业而导致缺勤的人群划分为“被临时裁员的失业者”。另外，如果受访者报告在一周普查期间没有工作，但预计会重新回到工作岗位，这类受访者也应该划为“被临时裁员的失业者”。将已经被解雇但预计会被重新雇佣的人员划为失业者（无论他们是否正在找其他工作）符合以往的惯例。
但总是会有复杂的情况出现。对临时缺勤人员的分类是否准确取决于缺勤的原因。如果因为恶劣天气、疾病、劳资纠纷、个人原因或休假等原因缺勤，这类人员应该被统计为就业者。这些理由似乎已经足够清晰。
以下是一些复杂的情况。如果缺勤人员因为“其他原因”缺勤，也有可能被统计为就业者。例如，他们可能请假照顾感染新冠肺炎的爱人。在引述“其他原因”时，普查人员通常会记录对原因的简要说明。劳工统计局在评估这些原因说明时，发现在3至6月的四次普查中，大量缺勤的受访者被错误分类成就业者，而不是失业人员或者被临时裁员。
在不进行季节性调整的基础上，经过修正错误分类误差，3月的总体失业率将提高约1个百分点，4月将提高约5个百分点，5月将提高约3个百分点，6月将提高约1个百分点。重要的是，最近几个月的错误分类误差变得更小，这让人们有理由相信失业率很快将不再受到这种误差的影响。
然而，过去四个月，错误分类误差对每个月失业率的影响程度各不相同。而且，按照笔者的理解，错误分类误差之大足以影响我们对数据的解读。5月总体失业率下降1.4%（从14.7%下降到13.3%），6月继续下降了2.2（下降到11.1%）。这意味着就业市场复苏的速度正在加快。
还有更好的消息。如图所示，修正错误分类误差之后，5月的失业率下降幅度更大，达到3.4%，远高于按照官方公布的总体失业率计算出的1.4%，6月修正后的失业率下降幅度为4.2%，而按照官方公布的总体失业率计算下降幅度只有2.2%。
很可惜，天下没有免费的午餐。图中还显示，与官方公布的总体失业率相比，修正误差之后，3月失业率的上涨幅度从0.9%提高到了1.9%，4月则从10.3%提高到14.3%。这四个月修正后的失业率都变得更高。4月的失业率达到19.7%，创下历史新高，比官方公布的总体失业率高出5个百分点，并且远远高于2009年10月大衰退期间的失业率最高峰10.0%。调整之后，美国就业市场的实际情况如何？经过修正之后，美国6月的失业率为12.1%，比官方公布的失业率高1个百分点。
图中还显示出疫情影响下就业市场的一个重要特征。失业率的波动性或者月度环比变化在持续加剧，比按照官方公布的失业率计算的波动更加明显。目前绝对是自二战以来美国就业市场最动荡的时期。6月的失业率单月下降4.2%，5月下降3.4%（经过修正后），都是自1948年以来的最大降幅。单月下降幅度排在第三位的是1949年11月，下降1.5%。4月至6月的失业率双月降幅为7.5%（修正后经过复合计算），是1948年以来的最大降幅。其次是1949年10月至12月，失业率双月降幅为1.3%。
更大的失业率波动令人担忧。这意味着一旦新冠疫情卷土重来，更多州将重新执行限制措施，比如关闭酒吧、健身房和餐厅等，因而失业率突然大幅反转的风险更高。与过去四个月总体失业率的影响相比，如果失业率再次攀升，就业市场和经济则可能遭到更大的破坏。劳工统计局报告称，随着各州纷纷重启经济，5月和6月的非农业就业人口总计增长750万人。一旦各州再次收紧防疫措施，这些就业岗位将面临风险。未来几个月，如果失业人数统计准确的话，不难想象，美国失业率肯定会超过4月的最高峰14.7%。劳工统计局在6月7日至6月13日进行了6月的家庭普查，但许多州在6月中下旬开始暂停经济重启，因此失业率已经开始攀升。
那些拒绝戴口罩或执行社交隔离的人，不仅让周围的市民面临感染新冠肺炎的风险，也会提高各州和地方政府被迫重新执行更严格的防疫措施的可能性，这将导致成千上万的雇主减少业务或者第二次停业，让成百上千万刚刚重回工作岗位的人再次承受失业的痛苦。（财富中文网）
本文作者约翰•D•芬尼特是福特汉姆大学（Fordham University）金融专业教授。
译者：Biz
美国总体失业率从5月的13.3%下降到6月的11.1%。这是好消息。今年4月，美国失业率达到14.7%，创自二战以来的新高。失业率连续下降意味着遭到公共卫生事件严重冲击的美国就业市场，似乎正在快速复苏。
此外，美国失业率好转的情况，比总体失业率所表现出来的趋势更令人高兴，这似乎也是一条好消息。然而不幸的是，失业率好转的原因是，上个月进行失业率对比的起点要比13.3%高很多。
美国劳工统计局（U.S. Bureau of Labor Statistics）在7月2日发布的新闻稿的脚注中写道，总体失业率可能低估了真实失业率，因为数据分类错误导致其每月家庭普查数据将临时失业者划分为就业者。劳工统计局在4月3日、5月8日和6月5日分别针对3月、4月和5月的就业情况发布的新闻稿中，也加了类似的脚注。
以下是对这项误差的解释。美国劳工统计局在每个月中旬进行家庭普查，利用普查结果计算出每个月的失业率，每个月结束后会很快对外公布。劳工统计局在脚注中解释说，其曾经指示普查人员将因为与疫情相关的企业临时停业而导致缺勤的人群划分为“被临时裁员的失业者”。另外，如果受访者报告在一周普查期间没有工作，但预计会重新回到工作岗位，这类受访者也应该划为“被临时裁员的失业者”。将已经被解雇但预计会被重新雇佣的人员划为失业者（无论他们是否正在找其他工作）符合以往的惯例。
但总是会有复杂的情况出现。对临时缺勤人员的分类是否准确取决于缺勤的原因。如果因为恶劣天气、疾病、劳资纠纷、个人原因或休假等原因缺勤，这类人员应该被统计为就业者。这些理由似乎已经足够清晰。
以下是一些复杂的情况。如果缺勤人员因为“其他原因”缺勤，也有可能被统计为就业者。例如，他们可能请假照顾感染新冠肺炎的爱人。在引述“其他原因”时，普查人员通常会记录对原因的简要说明。劳工统计局在评估这些原因说明时，发现在3至6月的四次普查中，大量缺勤的受访者被错误分类成就业者，而不是失业人员或者被临时裁员。
在不进行季节性调整的基础上，经过修正错误分类误差，3月的总体失业率将提高约1个百分点，4月将提高约5个百分点，5月将提高约3个百分点，6月将提高约1个百分点。重要的是，最近几个月的错误分类误差变得更小，这让人们有理由相信失业率很快将不再受到这种误差的影响。
然而，过去四个月，错误分类误差对每个月失业率的影响程度各不相同。而且，按照笔者的理解，错误分类误差之大足以影响我们对数据的解读。5月总体失业率下降1.4%（从14.7%下降到13.3%），6月继续下降了2.2（下降到11.1%）。这意味着就业市场复苏的速度正在加快。
还有更好的消息。如图所示，修正错误分类误差之后，5月的失业率下降幅度更大，达到3.4%，远高于按照官方公布的总体失业率计算出的1.4%，6月修正后的失业率下降幅度为4.2%，而按照官方公布的总体失业率计算下降幅度只有2.2%。
很可惜，天下没有免费的午餐。图中还显示，与官方公布的总体失业率相比，修正误差之后，3月失业率的上涨幅度从0.9%提高到了1.9%，4月则从10.3%提高到14.3%。这四个月修正后的失业率都变得更高。4月的失业率达到19.7%，创下历史新高，比官方公布的总体失业率高出5个百分点，并且远远高于2009年10月大衰退期间的失业率最高峰10.0%。调整之后，美国就业市场的实际情况如何？经过修正之后，美国6月的失业率为12.1%，比官方公布的失业率高1个百分点。
图中还显示出疫情影响下就业市场的一个重要特征。失业率的波动性或者月度环比变化在持续加剧，比按照官方公布的失业率计算的波动更加明显。目前绝对是自二战以来美国就业市场最动荡的时期。6月的失业率单月下降4.2%，5月下降3.4%（经过修正后），都是自1948年以来的最大降幅。单月下降幅度排在第三位的是1949年11月，下降1.5%。4月至6月的失业率双月降幅为7.5%（修正后经过复合计算），是1948年以来的最大降幅。其次是1949年10月至12月，失业率双月降幅为1.3%。
更大的失业率波动令人担忧。这意味着一旦新冠疫情卷土重来，更多州将重新执行限制措施，比如关闭酒吧、健身房和餐厅等，因而失业率突然大幅反转的风险更高。与过去四个月总体失业率的影响相比，如果失业率再次攀升，就业市场和经济则可能遭到更大的破坏。劳工统计局报告称，随着各州纷纷重启经济，5月和6月的非农业就业人口总计增长750万人。一旦各州再次收紧防疫措施，这些就业岗位将面临风险。未来几个月，如果失业人数统计准确的话，不难想象，美国失业率肯定会超过4月的最高峰14.7%。劳工统计局在6月7日至6月13日进行了6月的家庭普查，但许多州在6月中下旬开始暂停经济重启，因此失业率已经开始攀升。
那些拒绝戴口罩或执行社交隔离的人，不仅让周围的市民面临感染新冠肺炎的风险，也会提高各州和地方政府被迫重新执行更严格的防疫措施的可能性，这将导致成千上万的雇主减少业务或者第二次停业，让成百上千万刚刚重回工作岗位的人再次承受失业的痛苦。（财富中文网）
本文作者约翰•D•芬尼特是福特汉姆大学（Fordham University）金融专业教授。
译者：Biz
The U.S. headline unemployment rate dropped from 13.3% in May to 11.1% in June. This is very good news. Coming on the heels of the previous month’s drop from 14.7%, the highest unemployment rate since World War II, it suggests that the U.S. labor market, badly damaged by the coronavirus pandemic, appears to be on course for a quick recovery.
As good as the news seems to be, the improvement in the unemployment rate was even better than the headline numbers suggest. Yet more ominously, the improvement seems better because last month’s starting point for comparison was a lot higher than 13.3%.
The U.S. Bureau of Labor Statistics (BLS) reported in a footnote to its July 2 press release that the headline unemployment rate likely understates the true unemployment rate, owing to a misclassification error that counted some temporarily unemployed workers as employed in its monthly household survey data. The BLS included a similar footnote in its April 3, May 8, and June 5 press releases concerning the employment situation in March, April, and May, respectively.
Here is the explanation for the error. The BLS conducts a household survey in the middle of each month and uses the results to calculate the monthly unemployment rate, which it reports shortly after month-end. The BLS explained in a footnote that it had instructed its interviewers to classify people who were absent from work owing to temporary, coronavirus-related business closures as “unemployed on temporary layoff.” Accordingly, a survey respondent who reported she was not working during that week but said she expected to be summoned back to her job should have been recorded as “unemployed on temporary layoff.” This would be consistent with the usual practice of counting as unemployed people who have been laid off but expect to be recalled (regardless of whether they are looking for other work).
But, as often is the case, there is a complication. The correct classification of someone who is temporarily absent from her job depends on the reason for her absence. She would be counted as employed if she were temporarily absent because of bad weather, illness, a labor-management dispute, personal reasons, or vacation. These reasons seem clear enough.
Here is the complication. She can also be recorded as employed but absent from work for some “other reason.” For example, she might have taken a leave of absence to care for a loved one who was ill with COVID-19. When an “other reason” is cited, the interviewer usually records a brief description. When the BLS reviewed these descriptions, it found that a significant number of respondents had been misclassified in its four surveys for March through June as employed but absent from work, instead of unemployed on temporary layoff.
Correcting the misclassification error would have increased the headline unemployment rate by almost one full percentage point in March, almost five percentage points in April, about three percentage points in May, and about one percentage point in June, all on a not seasonally adjusted basis. Importantly, the misclassification errors are smaller in more recent months, which gives reason to hope they will soon stop affecting the unemployment rate.
Nevertheless, the errors affect the past four monthly unemployment rates to differing degrees. And the misclassification errors are large enough that they affect our interpretation of the data, as I will explain. The headline unemployment rate fell 1.4% in May (from 14.7% to 13.3%) and a further 2.2% in June (to 11.1%). This suggests an acceleration in the labor market’s recovery.
Here is even better news. As the accompanying chart illustrates, correcting the misclassification error reveals much larger-than-headline-rate decreases in May, a decrease of 3.4% versus only a 1.4% drop based on the headline unemployment rates, and in June, a decrease of 4.2% versus only a 2.2% drop implied by the headline unemployment rates.
Unfortunately, there is no free lunch here. The chart also reveals that correcting the errors indicates bigger-than-headline-rate increases in the unemployment rate for March, a 1.9% increase up from 0.9%, and for April, a 14.3% increase up from 10.3%. The corrected unemployment rates are higher for all four months. And the rate peaked at 19.7% in April, which is five percentage points above the headline rate. It is also far higher than the peak unemployment rate during the Great Recession of 10.0% in October 2009. Where do all the adjustments leave us? The corrections still leave the U.S. with a June unemployment rate of 12.1%, one percentage point above the headline number.
The chart also illustrates an important characteristic of the COVID-19 world. Volatility, in this case month-to-month changes in the unemployment rate, has increased and by even more than the headline numbers suggest. It is considerably more volatile than at any time during the post–World War II period. The one-month decreases of 4.2% in June and 3.4% in May (both on a corrected basis) are the biggest going back to 1948. The next largest one-month drop was 1.5% in November 1949. The two-month decrease of 7.5% (on a corrected basis and compounded) between April and June is the biggest going back to 1948. The next largest two-month drop was 1.3% between October and December 1949.
The greater unemployment rate volatility is concerning. It means there is a greater risk of a sharp reversal in the unemployment rate if a resurgence in the coronavirus leads more states to reimpose restrictions, such as closing bars, gyms, and restaurants. There is a risk of greater damage to the labor market and the economy than the past four months’ headline unemployment rates suggest. The BLS reported that total nonfarm payroll employment rose by 7.5 million in May and June combined as states reopened their economies. These jobs are potentially at risk if states retrench. And the U.S. unemployment rate could conceivably ratchet up above the April peak of 14.7%, especially if the number of unemployed are counted correctly in coming months. Since the June household survey was conducted between June 7 and June 13 and many states hit the pause button in the second half of June, unemployment may have already begun ticking up.
Those who refuse to wear masks or practice safe social distancing are not only placing many of their fellow citizens at risk of contracting COVID-19, but they are also increasing the likelihood that state and local governments will feel compelled to reimpose tighter restrictions that could force thousands of employers to cut back their operations or shut down their businesses a second time—causing millions of recently rehired workers to once again suffer the loss of their jobs.
John D. Finnerty is a professor of finance at Fordham University.
