我们的预测指数表明，美联储正在失去喘息空间。美国消费者价格，同比变化百分比，《经济学人》的预测模型。按价格变化排名对商品加权。整体通胀。按支出份额对商品加权。美联储在中期选举前的最后一次会议上维持利率不变。但美国经济分析局（BEA）10月29日发布的最新消费者价格数据显示，价格正在全面上涨。
Our predictive index suggests the Fed is running out of breathing room US consumer prices, , % change on a year earlier The Economist’s prediction model Weights items by price-change rank Headline inflation Weights items by share of spending The Federal Reserve held interest rates steady in its last meeting before America’s midterm elections. But the latest data on consumer prices released by the Bureau of Economic Analysis (BEA) on October 29th shows prices are rising across the board. After a relatively placid summer, the personal consumption expenditures (PCE) index—the Fed’s preferred measure of inflation—rose by half a percentage point in September alone.
在经历了一个相对平静的夏季之后，个人消费支出（PCE）指数——美联储偏好的通胀指标——仅9月一个月就上涨了0.5个百分点。尽管汽油价格上涨占了涨幅的近一半，但其余部分来自广泛的商品类别，其中以玩具和报纸为首。《经济学人》的预测通胀指数对新数据反应剧烈，从2.5%跃升至3.0%，这表明央行可能无法再等太久就要恢复其紧缩轨迹。
Although rising petrol prices accounted for nearly half the increase, the remainder came from a wide range of items, led by toys and newspapers. The Economist’s predictive inflation index reacted sharply to the new numbers. It jumped from 2.5% to 3.0%, suggesting that the central bank may not be able to wait much longer before resuming its tightening trajectory.
整体通胀数据往往难以指引价格走向。首先，被最广泛报道的指标——价格水平的同比变化——是一个滞后指标，它将一年前和上个月的数据视为同等重要。其次，通胀从来不是均匀的：某些商品和服务的价格总是比其他涨得更快。而某些商品价格的大幅变动可能对整体指数产生超大影响，即使它们反映的是较窄的趋势。
Headline inflation numbers are often a poor guide to where prices are heading. First, the most widely reported measure—the year-on-year change in the price level—is a lagging indicator that treats data from a year ago and last month as equally important. Second, inflation is never uniform: prices for some goods and services will always rise faster than others. And big changes in the prices of certain items can have an outsize effect on the overall index, even when they reflect narrower trends.
这种跳涨在食品和能源类别中最为常见，这也是它们被排除在“核心”通胀指标之外的原因。但价格冲击也可能影响其他商品。例如，2022年3月，随着新冠时期的旅行限制解除以及俄罗斯入侵乌克兰推高了航空燃油成本，机票价格上涨了11%。当月，该类别占整个经济通胀的八分之一。
Such jumps are most common in the food and energy categories, which is why they are excluded from “core” inflation measures. But price shocks can affect other items, too. For example, in March 2022 airfares rose by 11%, as covid-era travel restrictions were lifted and Russia’s invasion of Ukraine increased the cost of jet fuel. The category accounted for one-eighth of inflation across the entire economy that month.
《经济学人》设计了一个PCE指数的预测版本，旨在规避这些陷阱。它降低了价格涨跌剧烈项目的权重，并对足够广泛的短期通胀变化给予额外重视。上图中可见我方指数的预测性质：在通胀突然上升或下降的历史案例中，如1975年、2008年或2021年，我方序列比滞后的年度总体数字早几个月开始剧烈波动。目前我方指数比PCE同比3.7%的涨幅低0.7个百分点，这很好地表明当前数字在未来几个月可能会下降（但仍将维持在高位）。
The Economist has devised a predictive version of the PCE index that is designed to avoid these pitfalls. It gives less weight to items whose prices rise or fall rapidly, and places additional emphasis on short-term changes in inflation if they are sufficiently broad-based. The predictive nature of our index is visible in the chart above: in historical cases where inflation abruptly rose or fell, such as 1975, 2008 or 2021, our series starts to lurch months before the trailing annual headline figure does. At the moment our index is 0.7 percentage points below the 3.7% year-on-year change in PCE, a good indication that the current number is likely to fall in the coming months (but remain elevated).
下图追踪了自1960年以来美国通胀的集中度。当通胀高度集中——即少数昂贵项目以不同寻常的程度推高总体通胀率——这通常是美国动荡经济表面的泡沫，应预期通胀将回归其长期历史平均水平。另一方面，当通胀不集中——即通胀水平在比正常情况下更多的产品间相当均匀——通胀的近期变化往往具有“粘性”，并将持续到未来。
The chart below tracks concentration in American inflation since 1960. When inflation is exceedingly concentrated—meaning a handful of expensive items are driving the headline rate to an unusual degree—this is usually froth on America’s churning economy, and one should expect inflation to revert towards its longer-term historical average. On the other hand, when inflation is not concentrated—that is, the level of inflation is fairly even across more products than normal—recent changes in inflation tend to be “sticky”, and persist into the future.
例如，图表显示截至1999年1月的一年中存在高超额集中度。在香烟制造商同意就其产品导致的医疗费用达成昂贵的法律和解后，烟草价格上涨了33%，而仅有一个其他项目涨幅超过10%。2021年夏季，二手车、酒店客房和机票价格上涨推动了总体通胀。当前数据显示，在油价和燃料价格上涨导致5月短暂跳升后，超额集中度已趋于平稳。
For example, the chart shows that there was high excess concentration in the year to January 1999. The price of tobacco rose 33% after cigarette manufacturers agreed to an expensive legal settlement for medical costs caused by their products, whereas only one other item saw an increase above 10%. In the summer of 2021 rising prices for used cars, hotel rooms and airfares drove overall inflation. Current data show that excess concentration has levelled off after a brief jump in May following the rise in oil and fuel prices.
方法论 《经济学人》的通胀指数对美国经济分析局（BEA）的标准计算方法做了两处调整。第一，它对异常值进行折扣处理。与达拉斯联邦储备银行编制的衡量潜在通胀的“截尾均值”指标类似，我们按每个月价格变动幅度对PCE指数的数百个组成部分进行排序，并根据这些排名重新对每个月的数据加权。然而，与达拉斯联储直接剔除环比通胀最高和最低的项目不同，我们使用一条平滑曲线为每个项目赋予新的权重。
Methodology The Economist’s inflation index makes two adjustments to the BEA’s standard calculation. First, it discounts outliers. Just like the “trimmed-mean” measure of underlying inflation produced by the Federal Reserve Bank of Dallas, we sort the PCE index’s hundreds of components by their price changes in each month, and re-weight each month’s data based on those rankings. However, rather than outright discarding items with the highest and lowest month-on-month inflation like the Dallas Fed does, we assign a new weight to each item using a smooth curve.
在我们的指标中，月度价格变动接近中位数的项目所获权重比它们在PCE指数中高出约35%。相比之下，处于第10和第90百分位的项目权重减少30%至35%，而处于最外围极值的项目权重则缩小至几乎为零。
Those with monthly price changes near the median get around 35% more emphasis in our measure than they do in the PCE index. In contrast, those at the 10th and 90th percentiles get 30-35% less weight, and those at the outer extremes shrink to almost zero.
接下来，我们的指数将各月数据融合为一个平均值，其中最新数据的权重大于较早数据。由于每月对异常值降权带来了稳定性，我们得以对最新数据赋予很大的权重：六个月前30天的信息，其重要性大致相当于当月一天的信息。这些比例的选择，是为了使我们指数在预测随后一年PCE通胀时的准确性最大化。*为帮助解释我们的预测为何与整体PCE通胀不同，我们首先计算指数中每个单独组成部分同比价格变动的标准差——这是衡量各组成部分价格变动是集中还是分散的一个指标。
Next, our index blends each of these months of data into an average, with the latest numbers counting more than older ones. The stability introduced by down-weighting outliers every month enables us to put a large premium on the most recent figures: 30 days’ worth of information from six months ago matters roughly as much as one day in the current month does. These ratios were chosen to maximise our index’s accuracy at forecasting PCE inflation during the subsequent year.* To help explain why our forecast differs from headline PCE, we first calculate the standard deviation of the year-on-year price changes for each individual component of the index—a measure of how bunched together or spread out they are.
然后，我们衡量这一标准差与整体通胀水平之间的平均历史关系。最后，我们考察当前的标准差是否符合这一历史关系。我们将偏离趋势的情况称为“过度集中”。过度集中往往与我们的预测通胀指数是高于还是低于整体通胀保持一致。
We then measure the average historical relationship between this standard deviation and the overall inflation level. Finally, we ask whether the standard deviation is in line with this historical relationship. We call a deviation from the trend “excess concentration”. Excess concentration tends to track whether our predictive inflation index is above or below headline inflation.
由于我们的指数是过去12个月通胀指数的时间加权乘积——每个月都有其自身的项目价格变化排名——上述山形图并不反映我们追踪器的确切构成。相反，它通过将我们的加权方案应用于各商品或服务的同比价格变化，得出了一个近似值。
Because our index is the time-weighted product of the past 12 monthly inflation indices—each with its own ranking of price changes by item—the hill-shaped image above does not reflect the exact composition of our tracker. Instead, it produces a close approximation by applying our weighting scheme to year-on-year price changes in each good or service.
来源：美国经济分析局，《经济学人》 *供关注统计的读者参考：在该模型于2025年5月首次发布时，我们的模型在预测随后12个月总体PCE通胀方面的均方根误差为1.42个百分点。这与单纯用总体通胀预测自身的1.63个百分点，以及排除食品和能源价格的“核心”指标的1.65个百分点形成对比。自1978年1月（达拉斯联储修剪均值指标覆盖的首个月）以来，我们的预测性通胀指数的均方根误差为1.29个百分点，而一月修剪均值为1.40个百分点、六月修剪均值为1.36个百分点、12月修剪均值为1.44个百分点。
Sources: U.S. Bureau of Economic Analysis, The Economist *For statistically minded readers, at the time this model was first published in May 2025, our model’s root mean squared error at predicting headline PCE inflation during the subsequent 12 months was 1.42 percentage points. This compares with 1.63 points for simply using headline inflation to predict itself, and 1.65 points for the “core” measure that excludes food and energy prices. Since January 1978, the first month covered by the Dallas Fed’s trimmed-mean measure, our predictive inflation index’s root mean squared error is 1.29 percentage points, compared with 1.40 points for the one-month trimmed mean, 1.36 points for the six-month, and 1.44 points for the 12-month.