Wagner Piazza Gaglianone and Luiz Renato Lima, "Constructing Optimal Density Forecasts from Point Forecast Combinations", Journal of Applied Econometrics, Vol. 29, No. 4, 2014, pp. 736-757. We collect our data from the Federal Reserve Bank of Philadelphia's Real Time Data Set for Macroeconomists (RTDSM), as well as from the Survey of Professional Forecasters (SPF), both published by the Federal Reserve Bank of Philadelphia. The results presented in the paper are constructed by using Eviews (version 7.2). Approximate size of (zipped) files: 1,2 MB. Unzipped files: 4,5 MB All files are ASCII files in DOS format. They are zipped in the file gl-files.zip. Unix/Linux users should use "unzip -a". The files used in the paper are: code1.txt code2.txt code3.txt code4.txt code5.txt code6.txt code7.txt code8.txt code9.txt figure1_code.txt figure1_data.txt figure2_data.txt recursive_estimation.txt rolling_window_estimation.txt vintages.txt *** Instructions for replication of Tables and Figures of the paper *** Figure 1 (i) Create a new Eviews workfile (e.g., called Figure1.wf1) with monthly data, from 1985m1 to 2016m12, then, import data from file: figure1_data.txt and rename the imported time series as series cpi12m. (ii) Rename the file: figure1_code.txt as an Eviews code file (e.g., called figure1_code.prg). (iii) Run the previous code. (iv) Figure 1 is depicted in object "graph2" of the Eviews file. The blue line is the twelve-month inflation rate based on U.S. Consumer Price Index (CPI) data from Philadelphia Fed's Real-Time Data Set for Macroeconomists (RTDSM), seasonally adjusted (vintage CPI11M8). Inflation rate is calculated as log(Pt/Pt-12), in which Pt is the CPI index. The figure also shows the forecast conditional densities (gray area) based on a Quantile Autoregression model QAR(3) for k=250 out-of-sample trajectories (replications), with forecast horizons h=1,...,48. Estimation sample period: 1985m1:2011m6. The red lines represent the out-of-sample empirical quantiles {0.25; 0.50; 0.75}. Figure 2 Data comes from file: figure2_data.txt. Inflation is measured with GDP or GNP deflator, depending on data vintage, and the inflation rate is defined as the annualized quarterly rate, defined as 400*log(Pt+h/Pt+h-1), h = 1; 2; 3; 4; 8 and 12; in which Pt is the (real-time) GDP (or GNP) price index. We collect our data from the Federal Reserve Bank of Philadelphia's Real Time Data Set for Macroeconomists (RTDSM). Vintage 2012Q2 is used to compute the inflation rate, trend and gap displayed in figure 2. The inflation trend is computed from a moving average process based on the previous four years of the observed real-time inflation rate. ** Remaining Tables and Figures ** (i) Firstly, rename all the files labeled as code1.txt, code2.txt,..., code9.txt as files with extension .prg (to be used as Eviews code files) (ii) Open the file: recursive_estimation.txt as an Eviews file and save it as a workfile with extension .wf1 (iii) Rename the current workfile page as: quarterly (iv) Create a new workfile page, with quarterly data, from 1947Q1 to 2016Q4. Rename it as: Vintages_GDP_price_index (v) Based on this workfile page (Vintages_GDP_price_index) import data from file: vintages.txt (vi) From this workfile page (Vintages_GDP_price_index) copy the two series: gdp_price_p2011q2 and gdp_price_vintage (vii) Move to the workfile page "quarterly" and paste the two copied series as links (i.e., edit, paste special, as link, ok to all, overwrite object all) (viii) Save the Eviews workfile: recursive_estimation.wf1 (ix) Then, in order to replicate the results presented in Tables 3, 4 and 5, and Figures 3 and 4, one should sequentially run the codes: code1.prg, code2.prg,..., code9.prg. It is worth mentioning that the convention for the name of the models used throughout the codes include additional auxiliary models 17 to 24 (not reported in the paper, but used to generate the density combination scheme of Granger, i.e., model 28 of the paper). This way, throughout the codes, models 1-16 are the same models discussed in the paper. However, models i = {17,...,28} of the paper are labeled as models {i+8} all over the codes, due to the mentioned additional auxiliary models. In order to generate the results based on the "rolling window estimation", one must repeat the previous steps from (i) to (ix) but using the file: rolling_window_estimation.txt instead of recursive_estimation.txt Notice that in code1.txt, one must select the type of model estimation by using the following code line: scalar estimation = 1 (or 3) for recursive estimation (or rolling window estimation). Table 3 Results are presented in Eviews objects: z_coverage_rate_90 and z_cover_90_pvalue, in which rows represent forecast horizon (h) and columns represent the models (according to mentioned convention for the name of the models throughout the codes). Table 4 Results are presented in Eviews object: z_score, in which columns represent forecast horizon (h) and rows represent the models. Table 5 (only for rolling window estimation) Results are presented in Eviews objects: z_ag_stat and z_ag_pvalue, in which columns represent forecast horizon (h) and rows represent the models. Figure 3 (rolling window estimation) Results are presented in the following Eviews objects: QRW75_intercept_h3, QRW75_m9_h3, QRW75_m10_h3, QRW75_m11_h3, QRW75_m12_h3, QRW75_m13_h3, QRW75_m14_h3, QRW75_m15_h3, QRW75_m16_h3. Figure 4 (rolling window estimation) Results for quantile level 0.25 are presented in the following Eviews objects: QRW25_intercept_h3, QRW25_m9_h3, QRW25_m10_h3, QRW25_m11_h3, QRW25_m12_h3, QRW25_m13_h3, QRW25_m14_h3, QRW25_m15_h3, QRW25_m16_h3. For the median results (i.e., quantile 0.50) use similar objects with 50 instead of 25 in the name of the objects.