Written and maintained by CASRAI Editorial Board
Published
On this page
The Survey of Professional Forecasters (SPF) is a quarterly survey of macroeconomic forecasts for the United States, begun in 1968 and run since 1990 by the Federal Reserve Bank of Philadelphia. It publishes the mean and median forecasts of a panel of professional forecasters, the spread between them and each panelist’s own answers, so researchers can use it as a record of what forecasters expected at each date.
What is the Survey of Professional Forecasters?
The Philadelphia Fed describes the SPF as the oldest quarterly survey of macroeconomic forecasts in the United States. It began in 1968 as a joint effort of the American Statistical Association and the National Bureau of Economic Research (NBER), and the Philadelphia Fed took it over in 1990. The staff of the bank’s Real-Time Data Research Center produce it. For a researcher it is a survey instrument whose respondents are experts rather than a random sample of households, so it measures expert expectations, not the views of the general public.
The survey is not the same thing as other expectations surveys. The Philadelphia Fed lists the European Central Bank’s Survey of Professional Forecasters (for the euro area) and the Livingston Survey as related surveys, and an analyst should state which one is meant, because their panels, variables and calendars differ.
Who answers the survey and how are they chosen?
The panel consists of professional forecasters with long track records. Responses are anonymous: each panelist has a confidential identification number that stays the same across variables, so individual forecast paths can be followed over time, and names appear only in the acknowledgments. The panel is therefore a judgment-based panel, not a probability sample, and results describe this group of forecasters rather than all economists. The Q4 2026 release reports 32 forecasters.
When is the survey released?
The SPF has been quarterly since 1990:Q2. The questionnaire is sent on the day the Bureau of Economic Analysis publishes its first estimate of GDP for the previous quarter (see the BEA GDP page), responses are due about ten days later, and the results are published about three days after the deadline. The Fourth Quarter 2026 survey was released on 16 November 2026. Because every panelist answers within the same short window after the same GDP release, the forecasts share an information set, which is why the survey is used for real-time studies of forecast accuracy and bias (history summarised in Croushore and Stark, 2019).
What variables and files does the SPF publish?
The Philadelphia Fed organises the data as follows. The table lists only the groupings named on its survey page, with the statistics it offers for each.
| Download group | What it contains |
|---|---|
| Baseline variables | Mean, median, cross-sectional dispersion and individual forecasts |
| Probability variables | Forecast probabilities for ranges of outcomes, such as GDP growth and unemployment |
| Inflation forecasts | Short-term forecasts and a 10-year-ahead forecast for inflation |
| Anxious Index | Probability of a decline in real GDP in the quarter after the survey quarter |
| Annual historical values | Historical values for real GDP and the GDP price index by survey date |
| Documentation and errata | A 62-page documentation file (updated 14 August 2026) and an errata file |
The data table below the article compares the dates on which the survey’s design changed, which matters for anyone building a long time series.
Formula: the statistics the survey summarises
Each variable arrives as a set of individual forecasts. The mean, median and a dispersion measure summarise them. For N forecasts x1, …, xN sorted from smallest to largest:
- Mean: x̄ = (x1 + x2 + … + xN) / N, where x̄ is the average and N is the number of forecasters answering that question.
- Median: the middle value when N is odd, and the average of the two middle values when N is even.
- Sample standard deviation, one common dispersion measure: s = √[ Σ (xi − x̄)² / (N − 1) ], where xi is forecaster i’s value and s is in the same units as the forecasts.
The SPF publishes its own cross-sectional dispersion files; the formulas above show how a mean, a median and a spread are computed from individual forecasts and let you rebuild a summary from the individual-forecast file.
Worked example: reading the Q4 2026 release
Mean versus median, with the arithmetic shown
The following five numbers are illustrative values, not SPF data: 2.0, 2.2, 2.3, 2.4 and 3.1 (percent growth). The sum is 12.0, so the mean is 12.0 / 5 = 2.4. The middle value is 2.3, so the median is 2.3. The squared deviations from 2.4 are 0.16, 0.04, 0.01, 0 and 0.49, which total 0.70; divided by N − 1 = 4 gives 0.175, and the square root is s ≈ 0.418. The single high forecast (3.1) pulls the mean above the median, which is why the survey publishes both.
Changes between two surveys
The Fourth Quarter 2026 release (16 November 2026, 32 forecasters) reports medians against the previous survey. The change column below is the current value minus the previous value, computed here.
| Median forecast | Previous survey | Q4 2026 survey | Change |
|---|---|---|---|
| Real GDP growth, 2026:Q4 (annualized, %) | 2.3 | 2.3 | 0.0 |
| Unemployment rate, 2026:Q4 (%) | 4.3 | 4.2 | −0.1 |
| Payrolls, 2026:Q4 (thousands per month) | 66.4 | 90.0 | +23.6 |
| Headline CPI, 2026:Q4 (annualized, %) | 2.5 | 4.0 | +1.5 |
| Headline PCE, 2026:Q4 (annualized, %) | 2.6 | 3.5 | +0.9 |
| 10-year CPI, 2026-2035 (annual average, %) | 2.30 | 2.49 | +0.19 |
| 10-year PCE, 2026-2035 (annual average, %) | 2.20 | 2.23 | +0.03 |
The current-quarter inflation forecasts moved by more than a percentage point, while the 10-year forecasts moved by 0.19 and 0.03 percentage points: short-horizon forecasts are revised far more than long-horizon ones, a pattern a researcher should check rather than assume.
How to interpret mean, median, dispersion and probabilities
| Statistic | What it tells you | Reading guidance |
|---|---|---|
| Median | The middle forecast of the panel | Headline figure in the release; less sensitive to one extreme forecaster |
| Mean | The average forecast | Moves toward an outlier; compare with the median to see skew |
| Cross-sectional dispersion | How far forecasters disagree on one date | Wider means less agreement; it is not a measure of forecast error |
| Individual forecasts | One panelist’s path under a confidential ID | Needed for accuracy studies by forecaster; the ID is the same across variables |
| Probability variables | Mean probability placed on each outcome range | The Q4 2026 release shows 11 ranges for real GDP growth and 10 for unemployment |
Is the Survey of Professional Forecasters a recession forecast?
No. The survey does not ask forecasters for the probability of a recession. It asks for the probability of negative quarter-over-quarter real GDP growth, for the survey quarter and each of the following four quarters. The Anxious Index is that probability for the quarter after the survey quarter. In the Q4 2026 release the risk of a negative quarter for 2026:Q4 is 11.5%, against 20.0% in the previous survey, a drop of 8.5 percentage points (20.0 − 11.5). A recession is a multi-quarter episode judged on several indicators, so a single negative-quarter probability should not be reported as “the probability of a recession.”
Assumptions and common mistakes
- Treating the panel as a random sample. The panelists are chosen professionals,, so state the panel size reported for each survey date.
- Ignoring growth-rate conventions. Growth rates are discretely compounded and annualized, in percentage points. Since 1992:Q1 panelists may report levels or growth rates, and levels are converted to growth rates; between 70% and 80% of panelists reported levels for nominal GDP, the GDP price index and real GDP over 1992 to 2024.
- Mixing variable start dates. Headline CPI was added in 1981:Q3 and headline and core PCE in 2007:Q1, so a PCE series cannot start earlier than 2007:Q1 in the survey.
- Reading dispersion as uncertainty about the outcome. Disagreement between forecasters is a property of the cross-section, not of any one forecaster’s confidence; the probability variables are the survey’s direct measure of stated uncertainty.
- Using the latest numbers as the history. The survey records forecasts as made on each date; compare them with the data available at that time, as the real-time data literature does, not only with later revised values. Time-series work of this kind is covered in time series analysis in R.
How to cite the Survey of Professional Forecasters (APA)
APA 7 treats the release as a data set with the Federal Reserve Bank of Philadelphia as author; see the APA data set reference examples. Because the publisher and author are the same, the publisher element is omitted.
Reference list: Federal Reserve Bank of Philadelphia. (2026). Fourth quarter 2026 Survey of Professional Forecasters [Data set]. https://www.philadelphiafed.org/surveys-and-data/real-time-data-research/spf-q4-2026
In-text: (Federal Reserve Bank of Philadelphia, 2026). When reporting a result, name the statistic and the survey date: “The median forecast of 32 professional forecasters for 2026:Q4 real GDP growth was 2.3% (Federal Reserve Bank of Philadelphia, 2026).”
How to calculate the summary statistics in Excel, R or Python
Each snippet reproduces the illustrative five-forecast example above (mean 2.4, median 2.3, standard deviation 0.418). With the real survey, put the individual forecasts for one variable and one survey date in the vector or column. The SPF’s Excel downloads, such as the annual historical values file (annual-historical-values-by-survey-date-selected-variables.xlsx), can be opened with the same tools.
Excel (values in A2:A6):
=AVERAGE(A2:A6)
=MEDIAN(A2:A6)
=STDEV.S(A2:A6)
R:
x = c(2.0, 2.2, 2.3, 2.4, 3.1)
mean(x); median(x); sd(x)
# readxl::read_excel("annual-historical-values-by-survey-date-selected-variables.xlsx")
Python:
import statistics as st
x = [2.0, 2.2, 2.3, 2.4, 3.1]
print(st.mean(x), st.median(x), round(st.stdev(x), 3))
# pandas.read_excel("annual-historical-values-by-survey-date-selected-variables.xlsx")
For researchers building survey-based variables from public statistical releases, the Current Population Survey is a second U.S. source that pairs naturally with SPF unemployment forecasts, and the economics and finance hub lists related guidance.
Sources
- Federal Reserve Bank of Philadelphia. Survey of Professional Forecasters
- Federal Reserve Bank of Philadelphia. SPF FAQs
- Federal Reserve Bank of Philadelphia. Fourth Quarter 2026 Survey of Professional Forecasters
- Croushore, D., and Stark, T. (2019). Fifty Years of the Survey of Professional Forecasters. Economic Insights, Federal Reserve Bank of Philadelphia
- U.S. Bureau of Economic Analysis. Gross Domestic Product
- American Psychological Association. Data set references (APA Style)
Data: Survey of Professional Forecasters: design milestones and the dates they apply from
Compiled from the Federal Reserve Bank of Philadelphia SPF survey page and SPF FAQs: when the survey and its variables started and when reporting conventions changed, for users building long time series.
Period covered: 1968/2026. Coverage: United States.
| Milestone | Period | Detail |
|---|---|---|
| Survey begins | 1968 | Conducted by the American Statistical Association and NBER |
| Philadelphia Fed takes over | 1990 | Federal Reserve Bank of Philadelphia runs the survey |
| Quarterly series in current form | 1990:Q2 | Quarterly from 1990:Q2 to present |
| Headline CPI added | 1981:Q3 | Headline CPI forecasts enter the survey |
| Levels or growth rates accepted | 1992:Q1 | Panelists may report levels or growth rates; levels are converted to growth rates |
| Headline and core PCE added | 2007:Q1 | Headline and core PCE inflation forecasts enter the survey |
Sources
- Survey of Professional Forecasters (Federal Reserve Bank of Philadelphia) (accessed 2026-11-17)
- SPF FAQs (Federal Reserve Bank of Philadelphia) (accessed 2026-11-17)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







