"... In PAGE 20: ...erformance. The results are quite similar. The #0Crst regression I run is: PPS = #0B + #0C 1 ID RISK + #0C 2 SY S RISK + #0C 3 YEAR DUMMY + CONTROL+ #0F #283.21#29 Table4 reports the median regression result. I report three di#0Berent measures of risk esti- mated using the market model.... ..."

"... In PAGE 20: ... 16 Second, I add the control for #0Crm size to control for size-related heterogeneity. Table5 reports the OLS regression results for the same regression. Here, in addition to the control for #0Crm size and CEO e#0Bort productivity, I control for the year and #0Crm #0Cxed 16 When I employ #0Crm age and Tobin apos;s Q as proxy for CEO e#0Bort productivity respectively, or use all three proxies, the results are similar.... ..."

"... In PAGE 21: ... In tables 6 and 7, I redo the regression, replacing the measure of #0Crm-speci#0Cc and system- atic risks using the Fama-French model decomposition with up to 60 monthly observations before the current year. Table6 has no control variable, table 7 add the controls for #0Crm size and CEO e#0Bort productivity. The results are qualitatively similar.... ..."

"... In PAGE 15: ...ates averaged to 0.0245 for the 1979{82 period (compared to 0.0180 for the full sample). Table3 reports the full-sample maximum likelihood estimates of the param- eters involved in the two-factor speci cation (1){(3). The estimates of the ARCH and GARCH coe cients of the factor GARCH(1;1) processes imply that 1; 2; and (I? 1? 2) are positive de nite.... ..."

"... In PAGE 27: ... A comparison of gure 8 with gure 9 con rms the statement made earlier in this section that the BS-model approximates the SJD-model quite accurate for a high jump intensity (say, =100), given that the total volatility of the jump component is held xed. Table4 presents a sample of corresponding model prices for European calls written on the market index. The column with the header apos;BS apos; gives BS-value for short-term ( = 1=12, i.... ..."

"... In PAGE 17: ... Further details on the exact composition of our panel are available in Storesletten, Telmer, and Yaron (1998). In Table1 , row 1, we reproduce point estimates from our previous paper for the the following time series process, yit = git(yt) + uit (19) uit = zit + quot;it ; quot;it N(0; 2 quot;) (20) zit = zi;t?1 + it ; it N(0; 2 ) ; (21) where yit is the logarithm of the i apos;th household apos;s labor market endowment and git(yt) is the portion of yit comprising of aggregate shocks as well as deterministic components of household-speci c earnings such as unobservable ` xed e ects apos; and deterministic variation attributable to household age, education level and so on. In Storesletten, Telmer, and Yaron (1998) we discuss our particular parameterization of g (which follows closely a number of studies in the labor market dynamics literature), provide estimates and discuss how sensitive our results are to alternatives.... In PAGE 17: ... In Storesletten, Telmer, and Yaron (1998) we discuss our particular parameterization of g (which follows closely a number of studies in the labor market dynamics literature), provide estimates and discuss how sensitive our results are to alternatives. The rst row of Table1 shows that the autocorrelation coe cient is relatively large, at 0.935, and that the conditional standard deviation of the persistent shock process is roughly 90% larger than that of the transitory shocks.... In PAGE 18: ... First, we simply choose the variance of the distribution from which an agent draws their intercept term at birth so that the average, theoretical cross-sectional variance matches that of the data. Values represented by this procedure are reported in Table1 , row 2. The resulting age-pro le for cross-sectional variance is represented by the dotted line in Figure 1.... In PAGE 18: ... Loosely speaking, we chose to match the cross-sectional variation associated with the youngest age-cohort, to match the slope required to hit the variation associated with agents just ready to retire (the 60 year olds), and to match the curvature of the age-pro le. The values which result are reported in the third row of Table1 and the theoretical age-pro le is represented by the dashed line in Figure 1. Note that the implied value for is substantially higher, at 0.... In PAGE 19: ....S. females in 1991 and population growth is set to 1.0%. The process for idiosyncratic labor income, equation (2), is implemented as a discrete approximation to the autoregressive time series model and is parameterized using our point estimates from Table1 . In order to highlight the implications of the xed e ects, i, we begin by setting them to zero for our benchmark economy.... In PAGE 19: ...section 6.4) we allow them to be non zero and implement them as an i.i.d. two state bino- mial process, with variance chosen to match our estimates in Table1 . The age dependent intercept terms, h, are chosen so that the age-dependent mean of the logarithm of labor income in our theory matches our measure from the PSID.... In PAGE 30: ...escribed in sections 5.1 and 5.2. Speci cally, we modify the idiosyncratic risk process to include xed e ects according to the parameter values from the second line of Table1 . We then conduct several experiments designed to isolate pure risk sharing e ects (i.... In PAGE 43: ...7 Age Cross Sectional Variance The solid line represents estimates of the cross-sectional variance of PSID labor market in- come (inclusive of `transfers apos;), described in detail in Storesletten, Telmer, and Yaron (1997). The dash-dot line represents population moments associated with the time series process (19), evaluated at parameter estimates obtained by GMM (the rst row of Table1 ). The dotted line represents the incorporation of apos; xed e ects, apos; where we choose the variance of the distribution from which these parameters are drawn in order to match average disper- sion across ages (the second row of Table 1).... In PAGE 43: ... The dash-dot line represents population moments associated with the time series process (19), evaluated at parameter estimates obtained by GMM (the rst row of Table 1). The dotted line represents the incorporation of apos; xed e ects, apos; where we choose the variance of the distribution from which these parameters are drawn in order to match average disper- sion across ages (the second row of Table1 ). The dashed line incorporates xed e ects by choosing parameter values in order to match the initial cross sectional dispersion, and the slope and curvature of the age-pro le (the third row of Table 1).... In PAGE 43: ... The dotted line represents the incorporation of apos; xed e ects, apos; where we choose the variance of the distribution from which these parameters are drawn in order to match average disper- sion across ages (the second row of Table 1). The dashed line incorporates xed e ects by choosing parameter values in order to match the initial cross sectional dispersion, and the slope and curvature of the age-pro le (the third row of Table1 ). Speci cally, we choose 2 to match the initial variance, 2 to match the slope (or, equivalently, the end-point), and to match the curvature.... ..."

"... In PAGE 58: ...6.67/16.48); via a simple bootstrap test, we flnd that this is signiflcant (the test has a p- value indistinguishable from zero). Similarly, consulting Table12 for Bulgaria, we see that measurement error explains as much as 39.5 per cent of total Ligon-Schechter vulnerability, and is similarly signiflcant.... In PAGE 59: ...o 0.20, with a standard error of 0.0063, and hence highly signiflcant. As it happens, Table 7 still recommends the LB estimator in this instance. Thus, Table12 uses the LB estimator to estimate vulnerability for all the households in both samples. We decompose total vulnerability into several components for each household, as described in Section 3.... In PAGE 59: ... This ei- ther means that Bulgarians truly experience more unexplained idiosyncratic risk than do the Vietnamese, or that the Bulgarian data set has more measurement error. In Table12 we decompose vulnerability into poverty, aggregate risk, idiosyncratic risk, and unexplained risk, as in Ligon and Schechter (2003).18 Poverty is the largest single component of vulnerability.... ..."

"... In PAGE 17: ... After presenting our benchmark estimation results we discuss a number of modi#0Ccations intended to incorporate a non-degenerate distribution of initial conditions and#2For agent speci#0Cc #0Cxed e#0Bects. Table1 reports parameter estimates obtained using GMM in conjunction with the mo- ments in equations #2812#29. In Panel A we report estimates which constrain #1B H = #1B L and in Panel B we relax this restriction.... In PAGE 18: ... Our estimates incorporate this cross sectional evidence, but only implicitly #28via age-dependence in our moments#29. Our guess is that a more formal treatmentofhow inequality increases with age would generate an even higher value for the parameter #1A than we report in Table1 #28Storesletten, Telmer, and Yaron #281997#29 investigate this further#29. Finally, providing a frame of reference for our results on heteroskedastic idiosyncratic shocks over the business cycle is more problematic; aside from Heaton and Lucas #281996#29 we are not aware of comparable studies.... In PAGE 18: ... The largest increase associated with an economic downturn is roughly 12#25, which is associated with the recession in the early 1980 apos;s #28the same answer is obtained using the cross sectional standard deviation#29. At #0Crst blush, this might seem inconsistent with the values reported in Table1 , where we estimate the increase in the conditional standard deviation of the persistent shock, coincident with a downturn, to be on the order of 126#25. What apos;s going on, however, is that the larger number is associated with the conditional distribution of a given agent apos;s innovation, whereas the smaller number is much more closely associated with the unconditional distribution.... In PAGE 21: ...hose of U.S. females in 1991 and population growth is set to 1.0#25. The process for idiosyncratic labor supply, equation #282#29, is implemented as a discrete approximation to the autoregressive time series model and is parameterized using our point estimates from Table1 . The age dependentintercept terms, #14 h , are chosen so that, on aver- age, our theoretical age-earnings pro#0Cle matches that of the PSID.... In PAGE 21: ...Further details are provided in appendix B. This calibration of our theoretical income process, equation #282#29, su#0Bers from two poten- tial sources of discrepancy relative to the PSID-based estimates from Table1 . The #0Crst is the... In PAGE 22: ... More to the point, we #0Cnd that if we use our theoretical economy to generate an arbitrarily long sequence of 3 year panel data sets #28which corresponds to our PSID sampling method#29, the application of our GMM estimator #28section 3.2#29 to these data yields estimates which closely match those from Table1 ; our simulated point estimates are 0.... In PAGE 50: ... For our speci#0Ccation, if we ignore the transitory shocks, quot; it ,aswell as the terms which capture cross sectional variation due to age and education #28see equation #287#29 in section 3.2#29, then our estimation in Table1 boils down to a time series model of the residuals from a regression involving only year-dummyvariables. In a large cross section this will be, z it = log c it , ~ E t log c it : whichhave a cross sectional mean of zero, by construction, and a sample mean of zero, by least squares.... In PAGE 51: ...Table1 correspond to, z i;t+1 , z it =#281, #1A#29z it + #11 i;t+1 For values of #1A close to unity the variance of changes in zit is approximately equal to the variance of #11 i;t+1 . The left side of equation #2820#29 is, therefore, approximately equal to the variance of innovation, #11 i;t , ~ V t+1 #12 log c i;t+1 =c t+1 c it =c t #13 #19 ~ V t+1 #28#11 i;t+1 #29 In this sense our estimates of #1B H and #1B L provide estimates of the quantity needed to calibrate the Constantinides-Du#0Ee model.... In PAGE 51: ... Since aggregate consumption growth | the variable on the right side of equation #2820#29 | takes on only twovalues #283.8 percent and ,0:8 percent#29, computing the parameters a and b simply involves two linear equations: 0:037 = a +0:038b 0:181 = a , 0:008b; where the values on the left are the cross sectional variances from Table1 . The resulting values are a =0:156 and b = ,3:130.... ..."