The Limits of Prediction

Forecasting has become a vast industry, and its record is humbling. Economists failed to anticipate the 2008 crisis; public health models diverged wildly in 2020.

预测已经成为一个庞大的产业，而它的成绩单令人汗颜。经济学家没能预见到 2008 年的危机；2020 年的公共卫生模型之间差异极大。

Part of the difficulty is structural. Complex systems contain feedback loops in which predictions alter the behaviour being predicted. A widely published forecast changes what people do, and therefore changes the outcome.

部分困难是结构性的。复杂系统含有反馈回路，其中预测本身会改变被预测的行为。一份被广泛发布的预测会改变人们的做法，从而改变最终结果。

Another part is psychological. Confident narratives are more memorable than hedged probabilities, so the market rewards the wrong kind of accuracy. Pundits who are sometimes right and always certain flourish; those who are frequently right and openly uncertain do not.

另一部分是心理上的。自信的叙事比留有余地的概率更让人印象深刻，因此市场奖励的是错误的那种准确性。时而正确却始终笃定的评论家风生水起；经常正确却公开表示不确定的人则不然。

The productive response is not to abandon forecasting but to discipline it. Record predictions before the event, publish the errors, and treat uncertainty as information rather than weakness.

富有成效的回应不是放弃预测，而是对预测施加纪律：在事件发生前记录预测，公开错误，并把不确定性当作信息而非软弱。

There is a counter-example worth studying. A small group of forecasters consistently outperforms the rest, and their method is entirely unglamorous.

有一个值得研究的反例。一小群预测者的成绩持续优于其他人，而他们的方法毫无戏剧性。

They revise often, think in probabilities rather than in stories, and keep score. None of this makes prediction easy, but it makes prediction honest — and a forecast you can grade afterwards is useful even when it turns out to be wrong. It tells you something about the model, and something about the world.

他们频繁修正，用概率而不是故事来思考，并且记录自己的得分。这些都不会让预测变容易，但会让预测变得诚实 —— 而一条事后可以被打分的预测，即便最后是错的也有价值：它同时告诉了你一些关于这个模型、也关于这个世界的事情。