导读

当前全球经济环境充满不确定性使经济预测准确性和可靠性尤为关键。随着经济复杂性日益增加，银行预期偏误的问题也逐渐凸显。预期偏误不仅易导致预测失误，还可能引发一系列连锁反应，对经济政策制定和执行造成严重影响。因此，理解并有效应对这一心理倾向，已成为货币政策制定者必须重视的课题。

经济预测中的预期偏误现象并不罕见，甚至在某些情况下已成为一种普遍存在的问题。尤其在经济数据的分析和解读过程中，人们往往倾向于高估自身判断力，低估外部环境的复杂性和多变性。一个典型例子是在预测关键经济变量时，人们常常忽略潜在的误差幅度，过度依赖模型给出的精确数字。例如，在预测通胀率或失业率时，模型提供的预测区间常被视为高度可靠。然而，实际情况表明，这些区间往往过于狭窄，未能充分反映出模型的不确定性及外部变量的影响。过度自信的结果可能使政策制定者在面对实际经济变化时措手不及，从而导致错误的政策响应，进一步加剧经济波动。经济预测的复杂性还体现在长期趋势的判断上。经济模型通常基于历史数据进行推演，但由于经济环境的瞬息万变，这些预测往往难以准确反映未来的走势。类似于气象学家能够较为准确地预测短期天气变化，但在长期气候预测上存在较大不确定性，经济学家也常在短期经济波动的预测中表现较为准确，但对于长期趋势的判断则经常出现偏差。这一现象不仅反映了模型本身的局限性，也揭示了处理复杂系统时，过度自信可能导致严重后果。

为有效应对经济不确定性，货币政策制定者需要采用更为稳健和多元化的方法。一个值得借鉴的策略是，在制定政策时不仅依赖模型的预测结果，还应结合多种情景分析，评估可能的不同经济发展路径。通过这种方法，决策者能更好地识别潜在的风险和机会，避免因过度自信导致的决策失误。例如，在应对经济衰退的过程中，许多中央银行通过灵活调整利率政策，及时应对经济下行压力。这些政策的成功实施，很大程度上得益于决策者对不确定性的敏锐认知和应对能力。他们不仅依赖模型预测，还广泛收集和分析市场信息、企业反馈以及其他定性数据，以形成更为全面的经济判断。这种方法有效降低了过度自信带来的风险，确保了经济政策的稳健性和有效性。

与此相对的是，过度依赖单一预测模型常常使政策制定陷入困境。在某些情况下，决策者可能因为过分相信模型的精确度，而忽略潜在的巨大风险。以2008年全球金融危机为例，当时许多经济学家和政策制定者过于自信于现有经济模型的预测，低估了金融市场的不稳定性和复杂性。最终，正是这种过度自信导致了对金融风险的错误判断，致使全球经济陷入深度衰退。

因此，在制定经济政策时，特别是应对不确定性和风险时，保持谨慎和多样化的方法尤为重要。政策制定者需要保持开放的心态，不断学习并更新他们的经济判断，以应对不断变化的全球经济环境。通过将不同的模型和数据源相结合，决策者可以更全面地理解经济中的潜在风险，并制定出更为有效的政策，从而在复杂多变的经济形势中取得更好的政策效果。

作者 |安德鲁·豪斯（澳大利亚央行副行长）

Beware False Prophets

Andrew Hauser

Deputy Governor

Speech to the Economic Society of Australia (Queensland)

Brisbane – 12 August 2024

Introduction

My theme today is learning and the role it plays in monetary policymaking.

Overconfidence is not unique to economic commentary: it’s a universal human failing. In my remarks today, I want first to illustrate some real-world examples, before turning to ways in which central banks can avoid falling prey to it in our own deliberations. By forming contingent hypotheses about the future – rather than overly precise point forecasts. By learning continuously – from our own forecast errors, from diverse quantitative models, from corporate liaison and other qualitative intelligence-gathering, from experience in other countries, and from internal and external challenge, including scenarios and ‘what-ifs’. By communicating clearly and openly about what we don’t know, as well as what we do. And by adopting policy strategies that reflect risks to the outlook, as well as the central case. I will describe how some of these tools were applied in our most recent monetary policy round, and how we hope to develop them further.

Tackling uncertainty and overconfidence in central banking

So what, then, for monetary policymaking?

When we set interest rates, we have to look ahead – that is, make forecasts. That’s one source of uncertainty. Monetary policy works with long and variable lags – so our forecasts have to be medium-term, not short-term. That’s a second source of uncertainty. And most importantly of all, the things we are forecasting – inflation and unemployment – are the complex, time-varying outcomes of the decisions and interactions between many millions of people, companies and other organisations.

That puts us squarely in the world of Knightian uncertainty – of the State of Origin or Elon Musk. Unfortunately, most of the models used for economic analysis come from the worlds of paving stones and tomorrow’s weather! Estimate the average historical relationship between individual economic variables, run those relationships forward in time – and you get a set of deterministic point-estimate forecasts. Such model-based forecasts help show how the economy might respond if relationships remained exactly as they were in the past. But over-reliance on them has two key drawbacks

:

First, the probability of such paths being precisely correct is essentially zero. That makes them a poor basis for decision-making when used in isolation – because the absence of alternatives or fallbacks makes it harder to conceptualise other possible outcomes and weigh up the likelihood of those alternatives becoming reality. And that, in turn, can force forecasters either to underreact to news (forcing the data into their existing narrative, rather than learning) or to overreact (jettisoning one point-forecast for another demonstrably different path, which will also prove to be wrong).

Second, although it is possible to estimate confidence intervals around point forecasts using the models themselves, they are likely to be substantially too narrow (that is, overconfident) because they fail to account for the fact that the model may simply not be the right representation of reality.

So just as overconfidence can easily affect our everyday lives, it can affect monetary policymakers too. How do we guard against that?

The starting point is to avoid placing too much reliance on point forecasts in the first place, and instead frame our policy decisions in terms of contingent hypotheses or judgements. Some judgements may be strongly held, and hence given a high weight in the decision; others may be very tentative and given only a low weight. Both the hypotheses, and the weights attached to them, are continuously updated through a process of learning.

To bring this to life, consider a key question in the run-up to the RBA’s most recent policy decision:why has CPI inflation been so unexpectedly persistent?

A good place to start is to ask how unexpectedly persistent inflation has actually been – that is, tolearn from our own forecast errors. Such learning can be difficult for those who treat forecast ‘misses’ as failures. A more mature approach – and one long adopted by the RBA and other central banks – seeks instead to recognise that where forecasts are carefully constructed to make the best use of current data, ‘misses’ contain vital information about an intrinsically complex and stochastic world. This idea is also embedded in the ‘fancharts’ shown around the RBA’s outlook for inflation and other variables, which are calibrated using past forecast errors, not mechanical model outputs.

Three key facts emerge when we look at the recent path of forecast errors for inflation. First, although underlying inflation for the June 2024 quarter was broadly in line with our May forecast, inflation in earlier periods in2023–2024proved somewhat stickier than we had expected (Graph 4).

Second, the pattern for market expectations (derived from inflation swaps) has been somewhat similar, suggesting the upside surprises have not been limited to the RBA (Graph 5).

Third, some other jurisdictions have seen a rather greater incidence of inflationundershootsin recent quarters, raising the possibility that something slightly different might be underway in Australia. Graph6, for example, shows the picture for the euro zone.

To understand why past inflation outcomes may have been stronger than expected requires a hypothesis. In such circumstances, the economist’s reflex is to reach for a model: often some variant of a Phillips Curve, in which higher inflation reflects some combination of (a) higher expectedinflation, (b) higher demand, and (c) lower supply capacity. There is no evidence that longer term inflation expectations rose over this period in Australia. And GDP came in at or below forecast – which we tentatively assume is informative about demand conditions. We have therefore placed some weight on the possibility that past upside inflation surprises may have reflected somewhat weakersupplythan previously thought.

Now, it is one thing to hypothesise weaker supply, it is another to quantify it. And that’s because supply is not directly observable: it is a classic latent variable. So any estimate is subject to huge uncertainty. In the most recent monetary policy round, RBA staff approached this challenge in two ways.

The first was touse a range of alternative modelsthat posit different assumptions about the structure of the economy to estimate supply. It is common practice to transform the results of such models into estimates of ‘spare capacity’ in either the labour or product markets. Graph7 shows the range of spare capacity estimates for unemployment implied by the range of models. Choosing a single path within this range is subject to the very objections to point estimates that I just rehearsed. But to rationalise recent above-forecast inflation outcomes, the August 2024Statement on Monetary Policy(SMP) assumes that supply was somewhat weaker, and hence the labour market was somewhat tighter, than previously thought. And that extra weakness is assumed to persist, pushing up a little on the outlook for inflation. It must be said, however, that these changes in assumptions are tiny relative to the huge true range of uncertainty over these measures. So we have to be humble about our confidence in this judgement: spare capacity could easily be much higher, or much lower.

As a second cross-check, staff also compared their estimates to qualitative real-world indicators of capacity pressures from company surveys and the RBA’s liaison program.Graph8 shows that such measures lie at or above the upper end of our model-based estimates. That is also consistent with the messages of persistently elevated cost pressures I have personally heard from liaison visits to Townsville, Perth, Adelaide, Melbourne, Western and central Sydney in recent months.

Future inflation also depends on the outlook for demand – and here too we must guard against the risk of overconfidence. In the latest SMP, we do so in part by consideringalternative scenarios, drawing onlessons from overseas(particularly important for open economies such as Australia) and thepotential for, potentially sharp, changes in behaviour.

One scenario asks whether the unemployment rate may be about to pick up more rapidly than assumed, pulling down on demand and reducing inflationary pressure. In the central projection, employment growth only falls modestly, with total hours worked in the economy continuing to rise as the population and participation in the labour market grow. That is consistent with past downturns in activity of similar magnitude (Graph9).

But those assumptions could be wrong. What if unemployment rises more rapidly, as it has for example in Canada, Sweden and New Zealand (Graph 10)? Concerns that something similar might be about to happen in the United States caused last week’s sharp repricing in US rate expectations.

全文二维码

编译：浦榕

监制：崔洁

来源｜澳大利亚经济学会

版面编辑｜傅恒恒

责任编辑｜李锦璇、蒋旭

主编｜朱霜霜

近期热文

• 实习生招募 | 中国人民大学国际货币研究所（IMI）期待你的加入！
• 会议预告 | 铜钱铸造原料进口与清代的长崎贸易——以日方的统计为中心（9月12日）
• 可持续全球供应链与国际货币金融变革
• 日本存款保险制度与风险处置机制的改革与启示
• 政策性开发性金融工具：理论、案例与政策分析