Error in parallel

[Jose Fellmann]

Hi!
Very new to stata. I'm trying to replicate/extend the findings of this paper
The replication file is in there.


Imagine I have this data (this is just some randomly generated data. The actual data is too heavy to post)
* Generate random data similar to RegressionData.dta
clear
set seed 12345
set obs 90000 // Adjust to create 9000 unique banks with multiple time periods

* Simulate bank ID (rssdid) and time (date in quarterly format)
gen rssdid = floor((runiform() * 9000)) + 100000
gen date = tq(2000q1) + floor(runiform()*100)
format date %tq

* Drop duplicate combinations of bank ID and date
duplicates drop rssdid date, force

* Simulate interest income and expense rates
gen s1_intincrate = runiform(0.01, 0.06)
gen s1_intexprate = runiform(0.005, 0.05)

* Simulate NIM
gen s1_nim = s1_intincrate - s1_intexprate

* Simulate current and lagged federal funds rate shocks
gen s1_fedfunds = rnormal(0.02, 0.005)
gen l1_s1_fedfunds = s1_fedfunds + rnormal(0, 0.001)
gen l2_s1_fedfunds = l1_s1_fedfunds + rnormal(0, 0.001)
gen l3_s1_fedfunds = l2_s1_fedfunds + rnormal(0, 0.001)

* Simulate rank percentile and equity-to-assets ratio
gen rankpctile = runiform()*100
gen s1_equity_assets = runiform(0.05, 0.20)

* Set panel structure
xtset rssdid date

* Optional: sort and compress
sort rssdid date


My idea is to extend some results, and I'd like to parallelize the bootstrap, because otherwise, it takes too long.

[parallel setclusters 8
capture program drop t2_v3
program define t2_v3, rclass
preserve
* * * * *
bsample, cluster(date)
*sort rssdid date
*xtset rssdid date

// No time fixed effects
reghdfe s1_intexprate, absorb(i.rssdid i.rssdid#(c.s1_fedfunds c.l1_s1_fedfunds c.l2_s1_fedfunds c.l3_s1_fedfunds), save)
gen expbeta1 = __hdfe2__Slope1 + __hdfe2__Slope2 + __hdfe2__Slope3 + __hdfe2__Slope4
drop __hdfe1__-__hdfe2__Slope4

reghdfe s1_intincrate, absorb(i.rssdid i.rssdid#(c.s1_fedfunds c.l1_s1_fedfunds c.l2_s1_fedfunds c.l3_s1_fedfunds), save)
gen incbeta1 = __hdfe2__Slope1 + __hdfe2__Slope2 + __hdfe2__Slope3 + __hdfe2__Slope4
drop __hdfe1__-__hdfe2__Slope4

*reghdfe s1_equity_assets, absorb(i.rssdid i.rssdid#(c.s1_fedfunds c.l1_s1_fedfunds c.l2_s1_fedfunds c.l3_s1_fedfunds), save) --> for some reason, this does not work. the one below does
reghdfe s1_equity_assets, absorb(i.rssdid i.rssdid#c.s1_fedfunds i.rssdid#c.l1_s1_fedfunds i.rssdid#c.l2_s1_fedfunds i.rssdid#c.l3_s1_fedfunds, save)
gen equitybeta1 = __hdfe2__Slope1 + __hdfe3__Slope1 + __hdfe4__Slope1 + __hdfe5__Slope1
drop __hdfe2__Slope1 __hdfe3__Slope1 __hdfe4__Slope1 __hdfe5__Slope1

// With time fixed effects
reghdfe s1_intexprate, absorb(i.rssdid i.rssdid#(c.s1_fedfunds c.l1_s1_fedfunds c.l2_s1_fedfunds c.l3_s1_fedfunds) i.date, save)
gen expbeta2 = __hdfe2__Slope1 + __hdfe2__Slope2 + __hdfe2__Slope3 + __hdfe2__Slope4
drop __hdfe1__-__hdfe2__Slope4

reghdfe s1_intincrate, absorb(i.rssdid i.rssdid#(c.s1_fedfunds c.l1_s1_fedfunds c.l2_s1_fedfunds c.l3_s1_fedfunds) i.date, save)
gen incbeta2 = __hdfe2__Slope1 + __hdfe2__Slope2 + __hdfe2__Slope3 + __hdfe2__Slope4
drop __hdfe1__-__hdfe2__Slope4

reghdfe s1_equity_assets, absorb(i.rssdid i.rssdid#c.s1_fedfunds i.rssdid#c.l1_s1_fedfunds i.rssdid#c.l2_s1_fedfunds i.rssdid#c.l3_s1_fedfunds i.date, save)
gen equitybeta2 = __hdfe2__Slope1 + __hdfe3__Slope1 + __hdfe4__Slope1 + __hdfe5__Slope1
drop __hdfe2__Slope1 __hdfe3__Slope1 __hdfe4__Slope1 __hdfe5__Slope1

collapse incbeta1 expbeta1 incbeta2 expbeta2 equitybeta1 equitybeta2 rankpctile, by(rssdid)
winsor2 *beta*, cuts(5 95) replace
gen nimbeta1 = incbeta1 - expbeta1
gen nimbeta2 = incbeta2 - expbeta2
*pause
*gen niibeta = incbeta2 - expbeta2
// No time fixed effects
reg incbeta1 expbeta1
mat coef1 = e(b)
return scalar expbeta1_coef = el(coef1, 1, 1)
return scalar cons1_coef = el(coef1, 1, 2)
mat coef2 = e(V)
return scalar expbeta1_var = el(coef2, 1, 1)
return scalar cons1_var = el(coef2, 2, 2)
return scalar N1_100 = e(N)

reg incbeta1 expbeta1 if rankpctile >= 91
mat coef3 = e(b)
return scalar expbeta1_coef_10 = el(coef3, 1, 1)
return scalar cons1_coef_10 = el(coef3, 1, 2)
mat coef4 = e(V)
return scalar expbeta1_var_10 = el(coef4, 1, 1)
return scalar cons1_var_10 = el(coef4, 2, 2)
return scalar N1_10 = e(N)

reg incbeta1 expbeta1 if rankpctile >= 96
mat coef5 = e(b)
return scalar expbeta1_coef_5 = el(coef5, 1, 1)
return scalar cons1_coef_5 = el(coef5, 1, 2)
mat coef6 = e(V)
return scalar expbeta1_var_5 = el(coef6, 1, 1)
return scalar cons1_var_5 = el(coef6, 2, 2)
return scalar N1_5 = e(N)

reg incbeta1 expbeta1 if rankpctile >= 100
mat coef7 = e(b)
return scalar expbeta1_coef_1 = el(coef7, 1, 1)
return scalar cons1_coef_1 = el(coef7, 1, 2)
mat coef8 = e(V)
return scalar expbeta1_var_1 = el(coef8, 1, 1)
return scalar cons1_var_1 = el(coef8, 2, 2)
return scalar N1_1 = e(N)

// With time fixed effects
reg incbeta2 expbeta2
mat coef9 = e(b)
return scalar expbeta2_coef = el(coef9, 1, 1)
return scalar cons2_coef = el(coef9, 1, 2)
mat coef10 = e(V)
return scalar expbeta2_var = el(coef10, 1, 1)
return scalar cons2_var = el(coef10, 2, 2)
return scalar N2_100 = e(N)

reg incbeta2 expbeta2 if rankpctile >= 91
mat coef11 = e(b)
return scalar expbeta2_coef_10 = el(coef11, 1, 1)
return scalar cons2_coef_10 = el(coef11, 1, 2)
mat coef12 = e(V)
return scalar expbeta2_var_10 = el(coef12, 1, 1)
return scalar cons2_var_10 = el(coef12, 2, 2)
return scalar N2_10 = e(N)

reg incbeta2 expbeta2 if rankpctile >= 96
mat coef13 = e(b)
return scalar expbeta2_coef_5 = el(coef13, 1, 1)
return scalar cons2_coef_5 = el(coef13, 1, 2)
mat coef14 = e(V)
return scalar expbeta2_var_5 = el(coef14, 1, 1)
return scalar cons2_var_5 = el(coef14, 2, 2)
return scalar N2_5 = e(N)

reg incbeta2 expbeta2 if rankpctile >= 100
mat coef15 = e(b)
return scalar expbeta2_coef_1 = el(coef15, 1, 1)
return scalar cons2_coef_1 = el(coef15, 1, 2)
mat coef16 = e(V)
return scalar expbeta2_var_1 = el(coef16, 1, 1)
return scalar cons2_var_1 = el(coef16, 2, 2)
return scalar N2_1 = e(N)

// Now add the regressions of equity on nimbeta1
reg equitybeta1 nimbeta1
mat coef17 = e(b)
return scalar nimbeta1_coef = el(coef17, 1, 1)
return scalar cons3_coef = el(coef17, 1, 2)
mat coef18 = e(V)
return scalar nimbeta1_var = el(coef18, 1, 1)
return scalar cons3_var = el(coef18, 2, 2)
return scalar N3_100 = e(N)

reg equitybeta1 nimbeta1 if rankpctile >= 91
mat coef19 = e(b)
return scalar nimbeta1_coef_10 = el(coef19, 1, 1)
return scalar cons3_coef_10 = el(coef19, 1, 2)
mat coef20 = e(V)
return scalar nimbeta1_var_10 = el(coef20, 1, 1)
return scalar cons3_var_10 = el(coef20, 2, 2)
return scalar N3_10 = e(N)

reg equitybeta1 nimbeta1 if rankpctile >= 96
mat coef21 = e(b)
return scalar nimbeta1_coef_5 = el(coef21, 1, 1)
return scalar cons3_coef_5 = el(coef21, 1, 2)
mat coef22 = e(V)
return scalar nimbeta1_var_5 = el(coef22, 1, 1)
return scalar cons3_var_5 = el(coef22, 2, 2)
return scalar N3_5 = e(N)

reg equitybeta1 nimbeta1 if rankpctile >= 100
mat coef23 = e(b)
return scalar nimbeta1_coef_1 = el(coef23, 1, 1)
return scalar cons3_coef_1 = el(coef23, 1, 2)
mat coef24 = e(V)
return scalar nimbeta1_var_1 = el(coef24, 1, 1)
return scalar cons3_var_1 = el(coef24, 2, 2)
return scalar N3_1 = e(N)

// Now the same, but with time fixed effects
reg equitybeta2 nimbeta2
mat coef25 = e(b)
return scalar nimbeta2_coef = el(coef25, 1, 1)
return scalar cons4_coef = el(coef25, 1, 2)
mat coef25 = e(V)
return scalar nimbeta2_var = el(coef25, 1, 1)
return scalar cons4_var = el(coef25, 2, 2)
return scalar N4_100 = e(N)

reg equitybeta2 nimbeta2 if rankpctile >= 91
mat coef26 = e(b)
return scalar nimbeta2_coef_10 = el(coef26, 1, 1)
return scalar cons4_coef_10 = el(coef26, 1, 2)
mat coef27 = e(V)
return scalar nimbeta2_var_10 = el(coef27, 1, 1)
return scalar cons4_var_10 = el(coef27, 2, 2)
return scalar N4_10 = e(N)

reg equitybeta2 nimbeta2 if rankpctile >= 96
mat coef28 = e(b)
return scalar nimbeta2_coef_5 = el(coef28, 1, 1)
return scalar cons4_coef_5 = el(coef28, 1, 2)
mat coef29 = e(V)
return scalar nimbeta2_var_5 = el(coef29, 1, 1)
return scalar cons4_var_5 = el(coef29, 2, 2)
return scalar N4_5 = e(N)

reg equitybeta2 nimbeta2 if rankpctile >= 100
mat coef30 = e(b)
return scalar nimbeta2_coef_1 = el(coef30, 1, 1)
return scalar cons4_coef_1 = el(coef30, 1, 2)
mat coef31 = e(V)
return scalar nimbeta2_var_1 = el(coef31, 1, 1)
return scalar cons4_var_1 = el(coef31, 2, 2)
return scalar N4_1 = e(N)


* * * * *
restore
end

* * * * * * * * * * * * * * *
* B - Simulation
* * * * * * * * * * * * * * *
set seed 2020

parallel bs reps(10): simulate, expbeta1 = r(expbeta1_coef) var1 = r(expbeta1_var) cons1 = r(cons1_coef) consvar1 = r(cons1_var) N1_100 = r(N1_100) ///
expbeta1_10 = r(expbeta1_coef_10) var1_10 = r(expbeta1_var_10) cons1_10 = r(cons1_coef_10) consvar1_10 = r(cons1_var_10) N1_10 = r(N1_10) ///
expbeta1_5 = r(expbeta1_coef_5) var1_5 = r(expbeta1_var_5) cons1_5 = r(cons1_coef_5) consvar1_5 = r(cons1_var_5) N1_5 = r(N1_5) ///
expbeta1_1 = r(expbeta1_coef_1) var1_1 = r(expbeta1_var_1) cons1_1 = r(cons1_coef_1) consvar1_1 = r(cons1_var_1) N1_1 = r(N1_1) ///
expbeta2 = r(expbeta2_coef) var2 = r(expbeta2_var) cons2 = r(cons2_coef) consvar2 = r(cons2_var) N2_100 = r(N2_100) ///
expbeta2_10 = r(expbeta2_coef_10) var2_10 = r(expbeta2_var_10) cons2_10 = r(cons2_coef_10) consvar2_10 = r(cons2_var_10) N2_10 = r(N2_10) ///
expbeta2_5 = r(expbeta2_coef_5) var2_5 = r(expbeta2_var_5) cons2_5 = r(cons2_coef_5) consvar2_5 = r(cons2_var_5) N2_5 = r(N2_5) ///
expbeta2_1 = r(expbeta2_coef_1) var2_1 = r(expbeta2_var_1) cons2_1 = r(cons2_coef_1) consvar2_1 = r(cons2_var_1) N2_1 = r(N2_1) ///
nimbeta1 = r(nimbeta1_coef) var3 = r(nimbeta1_var) cons3 = r(cons3_coef) consvar3 = r(cons3_var) N3_100 = r(N3_100) ///
nimbeta1_10 = r(nimbeta1_coef_10) var3_10 = r(nimbeta1_var_10) cons3_10 = r(cons3_coef_10) consvar3_10 = r(cons3_var_10) N3_10 = r(N3_10) ///
nimbeta1_5 = r(nimbeta1_coef_5) var3_5 = r(nimbeta1_var_5) cons3_5 = r(cons3_coef_5) consvar3_5 = r(cons3_var_5) N3_5 = r(N3_5) ///
nimbeta1_1 = r(nimbeta1_coef_1) var3_1 = r(nimbeta1_var_1) cons3_1 = r(cons3_coef_1) consvar3_1 = r(cons3_var_1) N3_1 = r(N3_1) ///
nimbeta2 = r(nimbeta2_coef) var4 = r(nimbeta2_var) cons4 = r(cons4_coef) consvar4 = r(cons4_var) N4_100 = r(N4_100) ///
nimbeta2_10 = r(nimbeta2_coef_10) var4_10 = r(nimbeta2_var_10) cons4_10 = r(cons4_coef_10) consvar4_10 = r(cons4_var_10) N4_10 = r(N4_10) ///
nimbeta2_5 = r(nimbeta2_coef_5) var4_5 = r(nimbeta2_var_5) cons4_5 = r(cons4_coef_5) consvar4_5 = r(cons4_var_5) N4_5 = r(N4_5) ///
nimbeta2_1 = r(nimbeta2_coef_1) var4_1 = r(nimbeta2_var_1) cons4_1 = r(cons4_coef_1) consvar4_1 = r(cons4_var_1) N4_1 = r(N4_1) ///
,reps(10) : t2_v3


]

That code breaks by saying


"
Parallel Computing with Stata (by GVY)
Clusters : 8
pll_id : w8quls7pe1
Running at : C:\Users\josef\OneDrive - KU Leuven\PhD - Leuven\Banking project\EMPIRICAL\Call Reports Chicago Fed\DepositSensitivity
Randtype : datetime
Waiting for the clusters to finish...
cluster 0003 Exited with error -198- while running the command/dofile (view log)...
cluster 0005 Exited with error -198- while running the command/dofile (view log)...
cluster 0006 Exited with error -198- while running the command/dofile (view log)...
cluster 0007 Exited with error -198- while running the command/dofile (view log)...
cluster 0008 Exited with error -198- while running the command/dofile (view log)...
cluster 0001 Exited with error -198- while running the command/dofile (view log)...
cluster 0002 Exited with error -198- while running the command/dofile (view log)...
cluster 0004 Exited with error -198- while running the command/dofile (view log)...
--------------------------------------------------------------------------------
Enter -parallel printlog #- to checkout logfiles.
--------------------------------------------------------------------------------
"


any ideas?

Thanks!

[Felix Bittmann]

Did you click on "view log" and read the actually error message? What it is?
Does the program run fine on its own? Without using parallel? If so, does it run with the regular simulate command?
EDIT: note that this is also running an older version of parallel. To update, see: https://github.com/gvegayon/parallel

[Jose Fellmann]

The code runs for a single run, ie, no parallel. Can't access the logs, the code does not create them

Thanks!

[Felix Bittmann]

With the current version, the syntax should be:

parallel sim, exp(a=r(returnval)) reps(100) seed(1 2 3 4): t2_v3

[Jose Fellmann]

Hi Felix,

Thanks for all the help. The thing is now I get this error

invalid '-'
stata(): 3598 Stata returned error
parallel_export_programs(): - function returned error
parallel_write_do(): - function returned error
<istmt>: - function returned error


Any further ideas?

Thanks!