Stata Command: npss
Stata Command: npss.ado
The top left panel of the graph below draws a nonparametric deconvolution kernel estimation of the density of the permanent component of log earnings. The top right panel draws a nonparametric deconvolution kernel estimation of the density of the transitory component of log earnings. The bottom left panel draws the conditional auto-skedastic function of the permanent component of log earnings as a measure of heterogeneous earnings risk. The results are based on PSID for years 2006, 2008 and 2010 across the financial crisis. The conditional auto-skedastic function is decreasing, implying that richer individuals have less permanent earnings risks than poorer individuals (Botosaru and Sasaki, 2018). The graphs are automatically produced by the Stata command npss as follows:
Installation:
. ssc install npss
Usage:
. npss y2006 y2008, skedastic(y2010)
Help:
. help npss
Reference: Botosaru, I. and Y. Sasaki (2018) Nonparametric Heteroskedasticity in Persistent Panel Processes: An Application to Earnings Dynamics. Journal of Econometrics, 203 (2), pp. 283-296. Paper.
npss -- Executes nonparametric estimation of heteroskedastic state space models.
Syntax
npss y1 y2 [if] [in] [, skedastic(varname) tp1(real) tp2(real)]
Description
npss executes nonparametric estimation of conditionally heteroskedastic state space models based on Botosaru and Sasaki (2018). Consider a state space model yi,t = ui,t + vi,t, where yi,t is observed (e.g., earnings), ui,t is unobserved (e.g., permanent component of earnings), and vi,t is unobserved (e.g., transitory component of earnings), with the process ui,t = ui,t-1 + wit. Taking yi,1 and yi,2 as input, the command nonparametrically estimates and draws the density functions of ui,1 and vi,1. Taking yi,1, yi,2 and yi,3 as input, the command also nonparametrically estimates and draws the conditional skedastic function of ui,2 given ui,1, e.g., as a measure of heterogeneous risks in permanent component of earnings.
Options
skedastic(varname) tells the command to estimate the skedastic function of ui,2 given ui,1. The input in this option is y3, the observed variable in the third time period after the first two, y1 and y2. Not calling this option tells the command to estimate only the density functions of ui,1 and vi,1.
tp1(real) sets the scale-normalized tuning parameter for estimation of the density functions. The default value is tp1(4).
tp2(real) sets the scale-normalized tuning parameter for estimation of the skedastic function. The default value is tp2(2).
Examples
(y2006, y2008, & y2010: earnings in 2006, 2008, & 2010, respectively.)
Estimation of the density functions of u(2006) and v(2006), using y2006 and y2008 as input:
. use "example_2006_2008_2010.dta" . npss y2006 y2008
Estimation the conditional skedastic function of u(2008) given u(2006), in addition to the density functions of u(2006) and v(2006), using y2006, y2008 and y2010 as input:
. use "example_2006_2008_2010.dta" . npss y2006 y2008, skedastic(y2010)
Stored results
npss stores the following in r():
Scalars r(N) observations
Macros r(cmd) npss
Matrices r(U) density f(U) of U r(V) density f(V) of V r(S) conditional skedastic function sigma(U)
Reference
Botosaru, I. and Y. Sasaki. 2018. Nonparametric Heteroskedasticity in Persistent Panel Processes: An Application to Earnings Dynamics. Journal of Econometrics, 203 (2), pp. 283-296. Link to Paper.
Authors
Irene Botosaru, University of Bristol, Bristol, UK. Yuya Sasaki, Vanderbilt University, Nashville, TN.