A randomized Bundesbank quote pushed treated German forecasters' 2023 DAX forecasts 1.346 points above controls', with a significant response only among those who see inflation hurting stocks. Conrad, Schölkopf, Weber and Brückbauer offer convincing experimental evidence on how professionals revise forecasts. Their evidence on allocation is much weaker. A desk should bring the experimental result into forecast reviews and leave the allocation result out of its signals. Comparisons conditioned on narrative choice depend on self-selection, mainly by 153 respondents surveyed from 5 to 12 December 2022.
Rival channels for the same inflation path
The ZEW Financial Market Survey polls German financial market experts monthly. Among the 153 respondents to its December 2022 module, 42% work at banks and 11% at asset managers. The 99 who report their experience average 31 years. Euro-area inflation had peaked at 10.6% in October 2022. The following month, cross-sectional disagreement over 2023 inflation reached 1.72 percentage points, a level the authors describe as unprecedented in the survey.
The authors first gave 162 participants an open text box and asked how inflation would affect the DAX in 2023. Just 64 replied. Nearly all used present-value reasoning, though they disagreed over which term would move. For the discount-rate camp, inflation changes ECB policy: the equity response depends on whether easing or further hikes follow. The cash-flow camp instead asks whether DAX firms can pass on costs, widening margins or letting them compress. Smaller groups reasoned through the business cycle or expected no effect.
Those responses became six narratives, from which forecasters were asked to choose in December. Among the 136 who selected one, 56 chose easing inflation and looser policy ("inflation optimists"). Margin compression attracted 28, and successful pass-through 11. The remaining 36 chose persistent inflation and further hikes ("inflation pessimists"). Five selected no or a neutral effect. The authors combine them with the 17 who left the question unanswered as 'neutral'. Before treatment, inflation optimists expected a 6.807% DAX return in 2023. Inflation pessimists' forecasts were 2.505 points lower and margin pessimists' 2.834 lower, both at p<0.01.
In the full December sample, the slope relating expected returns to long-run inflation expectations is -0.554 (p<0.05). The three negative-sign narratives have a slope of -0.775 (n=81). Among pass-through and neutral forecasters, it is +1.054 (n=18, p<0.1). Pooling the December responses therefore weakens the negative slope without eliminating it. The positive slope appears only in that group of 18 December forecasters.
The quarterly panel tells a less tidy story. It covers February 2022 to November 2023, with an insignificant pooled inflation coefficient of -0.365 across 809 observations. For forecasters selecting a negative-sign narrative in both December 2022 and September 2023, the coefficient is -0.788. The residual group also has a negative, insignificant slope in this panel. The authors argue that an insignificant pooled coefficient can arise when camps with positive and negative relationships cancel each other out.
What changed after the Bundesbank president's 7% quote?
Random assignment gave 74 forecasters Bundesbank President Nagel's remark that 2023 inflation would probably have "a seven before the decimal point." Another 79 served as controls. The authors believe theirs is the first RCT with financial professionals. Treated respondents finished 0.390 points closer to 7% (p<0.05, n=121). That pull was significant only among forecasters whose 90% interval extended more than 2 points from 7% on at least one side.
The treated group's 2023 return forecasts finished 1.346 points above controls' (p<0.05, n=91). Winsorizing at 2.5% reduces significance to the 10% level. A hawkish-sounding quotation raising equity forecasts seems odd. The interaction helps explain it: forecasters with high long-run inflation priors took 7% as relief. The revision increases from an insignificant 0.410 points at the first quartile of long-run inflation expectations to 1.886 at the third.
The quotation had been public since November 8. Salience therefore matters at least as much as news in interpreting the experiment. More importantly, the signal was randomized; narrative choice was self-selected. The paper says that "only forecasters who believe in a negative relationship" revise. This rests on an interaction of 1.232 (p<0.01, n=74) for those groups. Growth optimists and neutral forecasters have an insignificant coefficient of 0.281, in a group the authors describe as only 17 forecasters. We did not find a test comparing the two coefficients. A null from 17 observations tells us little. Although the authors acknowledge the small 17-forecaster group, the introduction still says that only negative-narrative forecasters revise.
The measurement changes across treatment as well. Before treatment, values are interval midpoints; afterward, they are point forecasts. The authors disclose that they did not ask about long-run inflation again after treatment. Instead, they assume pass-through using the 0.587 coefficient relating short- to long-run expectations.
Treated forecasters' posterior inflation averaged 6.40%, versus 5.92% for controls. Their pre-treatment interval midpoints had averaged 6.700, compared with 6.430 for controls. Lower bounds were 5.263 against 4.644; upper bounds were 8.138 against 8.216. The paper describes the groups as balanced. Each forecaster's pre-treatment return and long-run inflation expectations enter the return regressions as controls. The test for movement toward 7% uses each respondent's own pre-to-post change. These choices limit the explanatory reach of the imbalance.
An 18.5-point DAX gap, on paper
Forecasters allocated an average 54.8% to equities when asked to divide a hypothetical €1m between the DAX and a 2% safe asset. Optimists put in 64.44%. Margin pessimists allocated 20.356 points less, and inflation pessimists 18.53 less. Narrative dummies alone produce an adjusted R² of 0.149.
The authors report that narratives retain relevance after controlling for pre-treatment return and inflation expectations. Within the negative-sign group, an extra point of expected long-run inflation accompanies 4.446 points less in the DAX, with return expectations held fixed.
These allocations are stated choices, deliberately unincentivized. Narrative choice was never randomized, and the paper gives no evidence that any camp allocates better. For what a single year can tell us, the mean expected DAX return was 5.272%; the index returned 20.26%.
Keep narrative labels in the forecast meeting
Persistence gives the result its practical use. December narratives continued to separate six-month DAX forecasts in every monthly wave from January to August 2023. Inflation-pessimist coefficients ranged from -3.387 to -5.992. By the paper's count, 43 of 105 forecasters gave the same answer in September 2023 as in December. That count includes respondents who skipped the question in both waves. Still, the authors find the mechanism stable for most forecasters. The channel lasted longer than the inflation call.
Anyone combining sell-side or internal forecasts should record the channel behind each analyst's reasoning. The experiment shifted one camp's return forecasts while leaving the other's statistically unchanged. The camps also have inflation-return slopes with opposite signs in the December cross-section.
We could not test any of this.
The respondent-level narratives, forecasts and treatment assignments are ZEW data covered by a non-disclosure agreement. Assigning narratives to news would measure something different. The paper's claim concerns belief formation in one episode within one German panel. On that claim, it delivers.