Stefan Hochguertel, "Precautionary Motives and Portfolio Decisions", Journal of Applied Econometrics, Vol. 18, No. 1, 2003, 61-77. The dataset is a subsample of the CentER Savings Survey, CentERdata, Tilburg University. At the time of writing, description and documentation of the data set is available at: http://center-ar.kub.nl/CSS There, access restrictions will be spelled out that determine the set of possible users of the original data set. The present paper uses six waves of the data, 1993-1998, and in principle, all eligible households. Data used in the estimation sample described below are transformations of the original data set. Also note that the household identifier used in the estimation sample does not coincide with the household identifier used in the original data. This replacement was required from the data proprietor in order to make the data available on the JAE Data Archive. Wprovide one data set allowing to replicate all estimates displayed in the tables. Since the number of observation differs across columns within each table and also across tables (and subtables), we include specification selectors. That is, for replication of a particular regression, select only those observations where the respective specification indicator takes the value 1. To extract the necessary variables, use the table below for selection of variables. Variables in the data set are ordered in the order displayed in the table. Short description of data set: Overall size: 1,652 Kbyte format: plain ASCII delimiter: comma # Variables: 65 # Records: 9,314 (1 record per row) Variable names are not provided. Data are presorted by households and within each household by sampling year (variables 16 and 17, below). All data are in the file sh-saving.data, which is an ASCII file in DOS format. It is zipped in the file shdata.zip. Unix users should use "unzip -a". The paper uses the following set of estimators: (1) Cross Sectional Tobit, with exponential heteroskedasticity (2-4) Cross Sectional CQR with quantiles = 0.5, 0.3, 0.1 (5) Random Effects Tobit, with exponential heteroskedasticity (6) Fixed Effects Tobit, MD estimator combining two-wave Honore estimates. We have three separate samples for (1,5), (2-4), and (6). The paper displays results in two tables. Table 1 contains the baseline specification, Table 2 contains (selected) results from sensitivity analyses. List of Specifications: 10: Baseline (Table 1) 21: Risk aversion and optimism (Table 2) 22: Labor supply flexibility (Table 2) 23: Health risk (Table 2) 24: Habits (Table 2) The following variables are used in these specifications (further particulars on the variables may be found in the text of the paper): SPECIFICATION CN TY DESCRIPTION 10 21 22 23 24 ------------------------------------------------------------------------- Specification Indicators 01 B 1= obs used in Table 1, CQR estimates (cols 2-4) x 02 B 1= obs used in Table 2.1, CQR estimates (cols 2-4) x 03 B 1= obs used in Table 2.2, CQR estimates (cols 2-4) x 04 B 1= obs used in Table 2.3, CQR estimates (cols 2-4) x 05 B 1= obs used in Table 2.4, CQR estimates (cols 2-4) x 06 B 1= obs used in Table 1, FE estimates (col 6) x 07 B 1= obs used in Table 2.1, FE estimates (col 6) x 08 B 1= obs used in Table 2.2, FE estimates (col 6) x 09 B 1= obs used in Table 2.3, FE estimates (col 6) x 10 B 1= obs used in Table 2.4, FE estimates (col 6) x 11 B 1= obs used in Table 1, ML estimates (cols 1,5) x 12 B 1= obs used in Table 2.1, ML estimates (cols 1,5) x 13 B 1= obs used in Table 2.2, ML estimates (cols 1,5) x 14 B 1= obs used in Table 2.3, ML estimates (cols 1,5) x 15 B 1= obs used in Table 2.4, ML estimates (cols 1,5) x Identifiers 16 I household ID x x x x x 17 I year x x x x x Endogenous variable 18 F share of clearly safe assets x x x x x Regressors 19 B 1= moderate income uncertainty, 5 years ahead [ref: low] x x x x x 20 B 1= high uncertainty, 5 yrs ahd [ref: low] x x x x x 21 B 1= expect income to stay const [ref: increase] x x x x x 22 B 1= expect income to decrease [ref: increase] x x x x x 23 FS log fin. assets, <= 10k Dfl x x x x x 24 FS log fin. assets, 10-100k Dfl x x x x x 25 FS log fin. assets, >100k Dfl x x x x x 26 F log net worth x x x x x 27 F log noncapital income x x x x x 28 D household marginal tax rate x x x x x 29 B 1= high-income oversample x x x x x 30 I age x x x x x 31 F age squared/100 x x x x x 32 F age cubed/1000 x x x x x 33 B 1= educ : secondary [ref: low] x x x x x 34 B 1= educ : vocational [ref: low] x x x x x 35 B 1= educ : university [ref: low] x x x x x 36 B 1= education : diploma x x x x x 37 B 1= female x x x x x 38 I number of adults x x x x x 39 I number of children aged 0-12 x x x x x 40 I number of children aged 13+ x x x x x 41 B 1= unemployed [ref: paid job] x x x x 42 B 1= retired [ref: paid job] x x x x 43 B 1= disabled [ref: paid job] x x x x 44 B 1= self-employed [ref: paid job] x x x x 45 B 1= other [ref: paid job] x x x x 46 B 1= year 1993 [ref: 1998] x x x x 47 B 1= year 1994 [ref: 1998] x x x 48 B 1= year 1995 [ref: 1998] x x x x x 49 B 1= year 1996 [ref: 1998] x x x x 50 B 1= year 1997 [ref: 1998] x x x x x 51 B 1= risk aversion: low [ref: high] x 52 B 1= risk aversion: medium [ref: high] x 53 B 1= risk aversion: don't know [ref: high] x 54 B 1= happy [ref: very happy] x 55 B 1= neither happy nor unhappy [ref: very happy] x 56 B 1= unhappy [ref: very happy] x 57 B 1= can take additional job [ref: none] x 58 B 1= can change hours worked [ref: none] x 59 B 1= both [ref: none] x 60 B 1= daily smoking: < 20 cigs. [ref: none] x 61 B 1= daily smoking: >=20 cigs. [ref: none] x 62 B 1= daily alcohol: > 4 drinks x 63 F body mass index x 64 B 1= strong habits [ref: none] x 65 B 1= weak habits [ref: none] x ------------------------------------------------------------------------- CN TY DESCRIPTION 10 21 22 23 24 SPECIFICATION LEGEND: CN: Column Number TY: Type B: Binary indicator (dummy variable), values 0 or 1 F: Float, continuous distribution FS: Float, linear spline segment I: Integer (nonnegative) D: Float with discrete distribution x: relevant for replication of results in respective specification : variable refers to head of household. All others refer to the entire household. Stefan Hochguertel Dept Economics European University Institute Via dei Roccettini, 9 50016 San Domenico di Fiesole (FI) Italy e-mail: Stefan.Hochguertel@iue.it date: Nov 15, 2001.