Andrew C. Chang and Trace J. Levinson, "Raiders of the Lost High-Frequency Forecasts: New Data and Evidence on the Efficiency of the Fed's Forecasting", Journal of Applied Econometrics, forthcoming. All files are zipped in the file cl-files.zip. Empty directories mentioned below are not included. ## Directory and File List data_dictionary.txt - explanation of high-frquency forecast data (or see the codebook at the end of this readme file). ### /ado/plus Contains packages for Stata. The master_public.do file will reset ado paths to this folder. Contains the packages and associated help files for: eret2, estadd, estout, estpost, eststo, esttab, ivreg2, ranktest ### /Code/public asymmetricloss.do - reestimates PAP specifications using Quad-Quad loss figures_forecastplots.do - graphs figures in paper master_public.do - runs everything regs_final_public.do - summary stats, regression estimation, outputs .tex files of tables to /results/latex/regs_final/public regs_final_public_allobs.do - regressions keeping all revisions, including those < 1bps regs_final_public_clusteredstderr.do - robustness checks using clustered by FOMC and Newey-West std errors regs_final_public_fixed_event.do - Nordhaus fixed event type autocorrelation tests SPFandHFcompare.do - compares forecasts of Survey of Professional Forecasters and high-frequency forecasts trendvsgrowth.do - tests for forecast smoothing UnivariatemodelandHFcompareGDP.do and UnivariatemodelandHFcompareInflation.do - Compare RMSEs of high-frequency forecasts to AR(1) and AR(4) models ### /Data/public hf_final_allobservations_public.dta and .csv - complete high-frequency dataset. Approximately 4,600 observations. hf_final_tbonly_public.dta and .csv - Greenbook/Tealbook dataset, used for Mincer & Zarnowitz regressions that only use Greenbooks in section 3.1 (section 2, equation 1 of preanalysis plan). Approximately 700 observations. ### /Data/Real-time data pconx_first_second_third.xlsx - Real-time core PCE inflation from real-time dataset for macroeconomists hosted by the Federal Reserve Bank of Philadelphia. Approximately 100 observations. routput_first_second_third.xlsx - Real-time GDP from real-time dataset for macroeconomists hosted by the Federal Reserve Bank of Philadelphia. Approximately 200 observations. ### /Data/SPF [Individual/Mean/Median]_[COREPCE/RGDP]_[Level/Growth] - Survey of Professional Forecaster data from FRB Philadelphia, representing individual level/mean/median forecasts for inflation/GDP. The number of observations varies by the release; about 70 for mean core PCE, about 200 for mean GDP, about 2,500 for individual core PCE, about 8,600 for individual GDP. SPFDates.csv and .txt - SPF release dates, one per quarter of data. ### /Data - Modified/HF Empty folder for intermediate data manipulation ### /Data - Modified/SPF Empty folder for intermediate data manipulation ### /Figures Empty folder to output of figures_forecastplots.do ### /Results #### /logs/final Empty folder for regression log output. #### /latex/regs_final/public Empty folder for latex output files from regs_final_public.do (PAP regressions), with a subfolder for summary stats #### /latex/nonpap_regs Five empty subfolders, allobs, asymmetricloss, clustered_std, fixed_even and revisions for associated results outputs ## Codebook for hf_final_allobservations, hf_final_tbonly_public year: year (yyyy) forecast was made. date: date (ddmmmyyyy) forecast was made. dateqtr: quarter of date (yyyyq#). obsdate: quarter that is being forecasted (dateqtr + horizon), string type. projqtr: quarter that is being forecasted (dateqtr + horizon), float type. fore_dist: the number of calendar days between the current forecast, for a particular macroeconomic variable-horizon, and the previous such forecast for a given FOMC cycle. This variable equals zero when, for a given macroeconomic variable-horizon, there was no other such forecast for a given FOMC cycle (this situation implies we only had the Greenbook for that variable-horizon for that FOMC cycle). Used in weighting regressions that used high-frequency data. fomc: the date of the first day of the next regularly scheduled FOMC meeting (ddmmmyyyy). varible: the macroeconomic variable being forecasted. source: the document type we used to record the forecast. we also record briefing tables and texts as pre-FOMC (the in-person forecast update just prior to a FOMC meeting) or bi-weekly (all other non-pre-FOMC briefings). horizon: forecast horizon in quarters, -1 = 1 quarter backcast, 0 = current quarter, etc.. Forecasts are indexed by horizon based on the next regularly scheduled FOMC date, not based on calendar quarters. rgdp_fore: real GDP growth forecasts, p.p., a.r. pcepilfe_fore: core PCE inflation forecasts, p.p., a.r. rgdp_3rd: BEA 3rd release of real GDP growth for the quarter being forecasted, p.p., a.r. pcepilfe_3rd: BEA 3rd release of core PCE inflation for the quarter being forecasted, p.p., a.r. fomcqtr: the quarter of the upcoming regularly scheduled FOMC meeting (yyyyq#). prevfomc: date of previous regularly scheduled FOMC (ddmmmyyyy). from_end: the number of forecasts for a particular variable-horizon-FOMC until the last forecast for that variable-horizon-FOMC, (i.e., =0 for the last forecast available before a FOMC). weeks: number of weeks until the first day of the next regularly scheduled FOMC meeting, rounded down. weeks01: indicator for whether it is less than 14 days until the first day of the next regularly scheduled FOMC meeting (i.e., I{weeks == 0 | weeks == 1}). rgdp_rev: revision (first difference) to real GDP growth forecast based on previous forecast for real GDP growth at the indicated horizon within a FOMC cycle. rgdp_rev == missing for the first forecast of real GDP growth for the indicated horizon for that FOMC cycle. rgdp_rev includes consecutive observations where the forecast does not revise. regression file (regs_final_public.do) excludes consecutive observations where the revision is less than one basis point to weed out thin trading issues. pcepilfe_rev: same as rgdp_rev for core PCE inflation. rgdp_err: real GDP growth forecast errors, 3rd release - forecast. pcepilfe_err: core PCE inflation forecast errors, 3rd release - forecast. zscore: rolling 71 day average of Bloomberg forecast errors, standardized by rolling 2 year standard deviation of Bloomberg forecast errors, NOT weighted by S&P 500 futures returns (news(tau) without r_i_tau in the summation formula). zscorexreturn: news(tau), S&P 500 return weighted rolling standardized sum of Bloomberg forecast errors.