Goldman Sachs: Systematic Bias in US Productivity Forecasts; AI May Boost Long-Term Growth

Overview

This report analyzes historical patterns in US nonfarm business labor productivity forecasts, pointing out systematic biases among professional forecasters. Based on Survey of Professional Forecasters (SPF) data, the report finds that forecasters react slowly to trend turning points and subsequently over-correct. Goldman Sachs forecasts average annual productivity growth of 2.3% for 2026-2030, higher than the market consensus of 2%, with the primary difference stemming from a more optimistic assessment of AI contributions.

Core views

Regarding forecast accuracy, the errors of professional forecasters are comparable to simple historical average models, with a mean absolute error of approximately 0.7 percentage points. Historical data shows that forecasters failed to timely capture key turning points such as the acceleration in the late 1990s, the slowdown in the mid-2000s, and the post-pandemic rebound. In terms of behavioral patterns, analysis of individual data from over 150 forecasters reveals a two-stage characteristic: initial anchoring to the 10-year moving average leads to underreaction, while later corrections involve excessive extrapolation of recent trends. This pattern is prevalent in forecasts for multiple macroeconomic variables. Amid current forecast divergence, most institutions' forecasts are close to the 10-year average of 2%, whereas Goldman Sachs' 2.3% forecast reflects more positive assumptions regarding AI adoption. Surveys indicate that the market consensus implies a slow AI adoption scenario (2.0%), while moderate/fast scenarios could reach 2.5%-3.5%.

Analysis framework

The report utilizes 30 years of historical data from the Survey of Professional Forecasters (SPF) to compare forecast errors against the performance of simple historical average models. Using an analytical method developed by Bordalo et al., it correlates the magnitude of forecast revisions with subsequent errors to validate the underreaction-overreaction pattern. The assessment of AI impact synthesizes the 2026 survey by Karger et al. and multiple forecast scenarios from the CBO, OECD, and others.

Methodology notes

• Event Gaming and Behavioral FinanceExpectation Gap/Expectation Management

  Forecasters' underreaction and over-correction to trend turning points

  When economic variables undergo trend changes, forecasters initially tend to maintain their original judgments (anchoring effect) and only over-correct once evidence becomes sufficient, leading to a systematic pattern in forecast errors. This psychological bias is prevalent in various macro forecasts, including productivity and inflation.

• Industry/Sector Analysis FrameworkPenetration S-curve

  Non-linear characteristics of AI technology adoption speed on productivity impact

  The contribution of new technologies to productivity depends on the speed of adoption, which may accelerate after a period of slow initial accumulation. The essence of current forecast divergence lies in different assumptions regarding the slope of the AI adoption curve, with slow/moderate/fast scenarios corresponding to annual contribution differences of 0.1-1.3 percentage points.

Key data

• Goldman Sachs 2026-2030 Forecast2.3%Average annual nonfarm business labor productivity growth
• Market Consensus Forecast≈2.0%Close to the 10-year moving average
• Mean Absolute Forecast Error0.7ppComparable between professional forecasters and historical average models
• AI Slow Adoption Scenario2.0%2026-2030 productivity growth forecast
• AI Fast Adoption Scenario3.5%2026-2030 productivity growth forecast

Impact & implications

The report alerts investors to systematic biases in macro forecasts; current conservative market expectations for productivity growth may underestimate the potential contribution of AI. If AI adoption proceeds faster than expected, actual productivity growth could exceed forecasts, impacting interest rate paths and asset pricing. Policymakers need to focus on how the speed of technology diffusion reshapes long-term growth potential.

What to watch

• Comparison of actual AI technology adoption progress against forecast scenarios
• Signals from quarterly nonfarm productivity data confirming trends
• Direction of revisions to post-2030 productivity forecasts by major institutions