Economic Forecasting宏观经济指标预测与研究Macro Indicators, Forecasts & Research
更新于 2026-09-12 09:30:45Updated 2026-09-12 09:30:45
产出模块Output module

工业生产Industrial Production

把工业生产拆成三个不同问题:当前已发布什么、下一个官方目标期的现时估计是什么、研究模型是否已经通过严格外样本门控。Separate three questions: what is currently released, what the next official target period is estimated to be, and whether a research model has passed strict out-of-sample gates.

三层工作面Three operating layers

Monitoring、Nowcasting、Forecasting 不共享同一个状态标签。Monitoring, nowcasting, and forecasting do not share one status label.

MonitoringMonitoring
最新首发月环比Latest first-release monthly growth
0.20%
观测期 2026-07-31Observed 2026-07-31
已记录Recorded
NowcastingNowcasting
下一目标期估计Next target estimate
0.14%
目标期 2026-08-31Target 2026-08-31
基线保护Baseline guarded
ForecastingForecasting
模型资格Model qualification
基线Baseline
服务模型 Baseline[rolling_median_12]Serving Baseline[rolling_median_12]
未证明优势No edge proven

目标合同与信息边界Target contract and information boundary

首发冻结表现和当前修订历史分开保存;缺少同版本分母时不构造首发目标。First-release frozen performance is kept separate from revised history; no first-release target is constructed without the same-vintage denominator.

目标变量Target工业生产月度环比对数增长Industrial production monthly log growth
单位Unitpercent_mom
预测期限Horizonone_month
首发口径Vintagealfred_initial_release
信息截止Information cutoff2026-09-12 00:00
数据状态Data status实时抓取

模型与回测资格Model and backtest qualification

候选模型必须同时通过全样本、近期窗口和发布日期感知的嵌套滚动验证。Candidates must pass full-sample, recent-window, and release-aware nested rolling validation.

全期 MASEFull-sample MASE
1.069
小于1才表示相对基线有优势Below 1 indicates an edge versus baseline
近期 MASERecent MASE
1.000
近期窗口独立报告Reported on a separate recent window
95%覆盖率95% coverage
95.8%
校准后的区间覆盖Calibrated interval coverage
研究候选Research candidate
NestedEnsemble[robust_state_space+ar]
未通过服务门控前只作研究记录Research record until it passes the serving gate

桥接输入审计清单Bridge-input audit checklist

这些变量是候选输入,不等于已经被模型使用或已经产生预测优势。每个值必须在预测时点已经发布。These are candidate inputs, not proof that the model used them or gained an edge. Each value must have been released by the forecast cutoff.

输入Input经济角色Role资格状态Qualification
制造业工时Manufacturing hours
Manufacturing_Hours
劳动投入Labor input需经发布日期对齐后进入候选回测Requires release alignment before candidate backtesting
产能利用率Capacity utilization
Capacity_Util
供给约束Supply constraint需经发布日期对齐后进入候选回测Requires release alignment before candidate backtesting
能源价格Energy prices
WTI_Oil
成本压力Cost pressure需经发布日期对齐后进入候选回测Requires release alignment before candidate backtesting
制造业信心Manufacturing confidence
Manufacturing_Confidence
调查信号Survey signal需经发布日期对齐后进入候选回测Requires release alignment before candidate backtesting
耐用品订单Durable-goods orders
Durable_Goods
需求管道Demand pipeline需经发布日期对齐后进入候选回测Requires release alignment before candidate backtesting

实现值、预测和基线Realized values, forecast, and benchmark

焦点窗口查看近期拐点;全部窗口查看疫情结构突变和长期误差。Use the focus window for recent inflections and the full window for the pandemic break and long-run error.

工业生产月度增长Industrial production monthly growth
Baseline[rolling_median_12]

边界与下一步Boundaries and next steps

工业生产覆盖制造业、采矿业和公用事业,但不能单独代表服务业需求或完整 GDP。Industrial production covers manufacturing, mining, and utilities, but does not represent services demand or total GDP on its own.

可以保留Keep

同一 vintage 的首发月环比、发布日期、区间校准和基线保护。Same-vintage first-release growth, release dates, calibrated intervals, and baseline guard.

必须复核Review

桥接输入的真实可得日、缺失值与极端时期残差。Actual availability dates, missing values, and crisis-period residuals for bridge inputs.

禁止发布Do not publish

没有通过外层嵌套回测的复杂模型优势,或任何跨 vintage 的冻结准确率。Complex-model edge without nested outer validation, or any cross-vintage frozen accuracy claim.