/*
libname research 'L:\faculty\PHDShared\green\research\papers';
*set up 10 years 1000 firms;
*this file just starts with 10000 observations with one variable numbered 1 to 10000;
data firmyr;
	set research.id10000;
	do j=1 to 1000;
		if (j-1)*10<id<=j*10 then firm=j;
	end;
	drop j;
run;
data firmyr;
	set firmyr;
	year=id-(firm-1)*10;
	run;
*initial sales demand for each firm is distributed randomly around 100 with 50std
	the probability of the good state is randomly distributed around .5 and 
		stays the same througout the sample unless news is given that it is 
		the good or the bad state
	the firm's fixed costs are randomly imposed;
data firmyr;
	set firmyr;
	if year=1 then do;
		saled=100+50*rannor(0);
		pprg=.5 +.1*rannor(0);
	end;
	if year=1 and firm<=500 then do;
			fc=50+10*rannor(0);
			end;
		else do;
			fc=0;
		end;
	run; 
*subsequent demand is determined by state,
	the mean shifts by 25 up or down depending on the state and then there is a random
	growth term;
proc sort data=firmyr;
	by firm year;
	run;
data firmyr2;
	set firmyr;
	lagpprg=lag(pprg);
	lagsaled=lag(saled);
run;
data firmyr2;
	set firmyr2;
	if pprg eq . then pprg=lagpprg;
	saledg=lagsaled+25+(lagsaled*.05)*rannor(0);
	saledb=lagsaled-25+(lagsaled*.05)*rannor(0);
	if year=1 then saledg=saled;
	if year=1 then saledb=saled;
run;
data firmyr2;
	set firmyr2;
	drop lagsaled lagpprg;
	run;
options nonotes;
%macro lagsim;
	%do i=2 %to 10000;
data firmyr2;
	set firmyr2;
	if year ne 1 then lagpprg=lag(pprg); else lagpprg=.;
	if year ne 1 then lagsaledg=lag(saledg);else lagsaledg=.;
	if year ne 1 then lagsaledb=lag(saledb);else lagsaledb=.;
run;
data firmyr2;
	set firmyr2;
	if pprg eq . then pprg=lagpprg;
	if saledg=. then saledg=lagsaledg+25+(lagsaledg*.02)*rannor(0);
	if saledb=. then saledb=lagsaledg-25+(lagsaledg*.02)*rannor(0);
run;
data firmyr2;
	set firmyr2;
	drop lagpprg lagsaledg lagsaledb;
	run;
%end;
%mend lagsim;
%lagsim;
options notes;
data firmyr2;
	set firmyr2;
	drop saled;
	run;

*fc in any period determines capacity constraint: 
	the firm can sell three times as much as its fixed costs,
	if current demand is greater than capacity and the future expected avg demand
	is greater than capacity, in the next period, increase capacity to meet 
	future expected average demand and is available in the next period but current period cannot meet sales
	demand;
*first get expected future saled;
proc sort data=firmyr2;
	by firm descending year;
	run;
data firmyr3;
	set firmyr2;
	lagsaledg=lag(saledg);
	lag2saledg=lag2(saledg);
	lag3saledg=lag3(saledg);
	lag4saledg=lag4(saledg);
	lag5saledg=lag5(saledg);
	lag6saledg=lag6(saledg);
	lag7saledg=lag7(saledg);
	lag8saledg=lag8(saledg);
	lag9saledg=lag9(saledg);

	lagsaledb=lag(saledb);
	lag2saledb=lag2(saledb);
	lag3saledb=lag3(saledb);
	lag4saledb=lag4(saledb);
	lag5saledb=lag5(saledb);
	lag6saledb=lag6(saledb);
	lag7saledb=lag7(saledb);
	lag8saledb=lag8(saledb);
	lag9saledb=lag9(saledb);
	run;
data firmyr3;
	set firmyr3;
	*retain eavgsaled;
	if year=10 then eavgsaled=0;
	else if year=9 then eavgsaled=(pprg*lagsaledg + (1-pprg)*lagsaledb);
	else if year=8 then eavgsaled=((pprg*lagsaledg+lag2saledg)/2 + (1-pprg)*(lagsaledb+lag2saledb)/2);
	else if year=7 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg)/3 + (1-pprg)*(lagsaledb+lag2saledb+lag3saledb)/3);
	else if year=6 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg+lag4saledg)/4 + (1-pprg)*(lagsaledb+lag2saledb+lag3saledb+lag4saledb)/4);
	else if year=5 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg+lag4saledg+lag5saledg)/5 + (1-pprg)*(lagsaledb+lag2saledb+lag3saledb+lag4saledb+lag5saledb)/5);
	else if year=4 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg+lag4saledg+lag5saledg+lag6saledg)/6 + (1-pprg)*(lagsaledb+lag2saledb+lag3saledb+lag4saledb+lag5saledb+lag6saledb)/6);
	else if year=3 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg+lag4saledg+lag5saledg+lag6saledg+lag7saledg)/7 + (1-pprg)*(lagsaledb+lag2saledb+lag3saledb+lag4saledb+lag5saledb+lag6saledb+lag7saledb)/7);
	else if year=2 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg+lag4saledg+lag5saledg+lag6saledg+lag7saledg+lag8saledg)/8 + (1-pprg)*(lagsaledb+lag2saledb+lag3saledb+lag4saledb+lag5saledb+lag6saledb+lag7saledb+lag8saledb)/8);
	else if year=1 then eavgsaled=((pprg*lagsaledg+lag2saledg+lag3saledg+lag4saledg+lag5saledg+lag6saledg+lag7saledg+lag8saledg+lag9saledg)/9+ (1-pprg)*(lagsaledb+lag2saledb+lag3saledb+lag4saledb+lag5saledb+lag6saledb+lag7saledb+lag8saledb+lag9saledb)/9);
	run;
proc sort data=firmyr3;
	by firm year;
	run;
data firmyr4;
	set firmyr3;
	retain fc;
		do i=1 to 5000;
			if id=i and firm=lag(firm) then do;
				*capacity constraint lag(fc)*3;
				if lag(eavgsaled)>lag(fc)*3 then do;
					fc=lag(eavgsaled)/3;
					end;
				else do;
					fc=lag(fc);
					end;
			end;
		if saledg>3*fc then saleg=3*fc; else saleg=saledg;
		if saledb>3*fc then saleb=3*fc; else saleb=saledb;
		end;
	drop i;
run;	

*get average total cost for similar expected sales demand to create variable costs percentage;
proc sql;
	create table avgcostsf
	as select *,firm,sum(fc)/sum(eavgsaled) as avgpctcosts,avg(eavgsaled) as avgsaled
	from firmyr4
	group by firm;
	quit;
data fcm;
	set avgcostsf;
	where id<=5000;
	mavgsaled=avgsaled;
	keep avgpctcosts mavgsaled mergedum;
	mergedum=1;
	run;
data vcm;
	set avgcostsf;
	where id>5000;
	drop avgpctcosts;
	mergedum=1;
	run;
proc sql;
	create table m
	as select * 
	from fcm a,vcm b
	where a.mergedum=b.mergedum;
	quit;
data m;
	set m;
	avgsaleddif=abs(avgsaled-mavgsaled);
	run;
proc sort data=m;
	by firm year avgsaleddif;
	run;
proc sort data=m nodupkey;
	by firm year;
	run;
data m;
	set m;
	vcp=avgpctcosts;
	run;
data vc;	
	set m;
	saleg=saledg;
	saleb=saledb;
	drop avgpctcosts avgsaleddif mavgsaled mergedum;
	run;
data fc;
	set firmyr4;
	where id<=5000;
	vcp=0;
	run;
data firmyr5;
	set vc fc;
	drop avgsaled;
	run;
data firmyr5;
	set firmyr5;
	nig=saleg-saleg*vcp-fc;
	nib=saleb-saleb*vcp-fc;
	run;
*now present value the net income;
data firmyr5;
	set firmyr5;
	pvnig=nig/(1+.05)**year;
	pvnib=nib/(1+.05)**year;
	run;
proc sort data=firmyr5;
	by firm descending year;
	run;
data firmyr5;
	set firmyr5;
	lagpvnig=lag(pvnig);
	lag2pvnig=lag2(pvnig);
	lag3pvnig=lag3(pvnig);
	lag4pvnig=lag4(pvnig);
	lag5pvnig=lag5(pvnig);
	lag6pvnig=lag6(pvnig);
	lag7pvnig=lag7(pvnig);
	lag8pvnig=lag8(pvnig);
	lag9pvnig=lag9(pvnig);

	lagpvnib=lag(pvnib);
	lag2pvnib=lag2(pvnib);
	lag3pvnib=lag3(pvnib);
	lag4pvnib=lag4(pvnib);
	lag5pvnib=lag5(pvnib);
	lag6pvnib=lag6(pvnib);
	lag7pvnib=lag7(pvnib);
	lag8pvnib=lag8(pvnib);
	lag9pvnib=lag9(pvnib);
	run;
*get price;
proc sort data=firmyr5;
	by firm descending year;
	run;
libname research 'L:\faculty\PHDShared\green\research\papers';
data research.firmyr5;
	set firmyr5;
	run;
data firmyr6;
	set firmyr5;
	retain priceyrende;
	if year=10 then priceyrende=0;
	if year=9 then priceyrende=(pprg*lagpvnig + (1-pprg)*lagpvnib);
	if year=8 then priceyrende=(pprg*(lagpvnig+lag2pvnig) + (1-pprg)*(lagpvnib+lag2pvnib));
	if year=7 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig) + (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib));
	if year=6 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig) + (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib));
	if year=5 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig) + (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib));
	if year=4 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig) + (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib));
	if year=3 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig+lag7pvnig) + (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib+lag7pvnib));
	if year=2 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig+lag7pvnig+lag8pvnig) + (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib+lag7pvnib+lag8pvnib));
	if year=1 then priceyrende=(pprg*(lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig+lag7pvnig+lag8pvnig+lag9pvnig)+ (1-pprg)*(lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib+lag7pvnib+lag8pvnib+lag9pvnib));

	retain priceyrendg;
	if year=10 then priceyrendg=0;
	if year=9 then priceyrendg=(lagpvnig);
	if year=8 then priceyrendg=((lagpvnig+lag2pvnig));
	if year=7 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig));
	if year=6 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig));
	if year=5 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig));
	if year=4 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig));
	if year=3 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig+lag7pvnig));
	if year=2 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig+lag7pvnig+lag8pvnig));
	if year=1 then priceyrendg=((lagpvnig+lag2pvnig+lag3pvnig+lag4pvnig+lag5pvnig+lag6pvnig+lag7pvnig+lag8pvnig+lag9pvnig));

	retain priceyrendb;
	if year=10 then priceyrendb=0;
	if year=9 then priceyrendb=(lagpvnib);
	if year=8 then priceyrendb=((lagpvnib+lag2pvnib));
	if year=7 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib));
	if year=6 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib));
	if year=5 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib));
	if year=4 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib));
	if year=3 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib+lag7pvnib));
	if year=2 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib+lag7pvnib+lag8pvnib));
	if year=1 then priceyrendb=((lagpvnib+lag2pvnib+lag3pvnib+lag4pvnib+lag5pvnib+lag6pvnib+lag7pvnib+lag8pvnib+lag9pvnib));
run;
proc sort data=firmyr6;
	by firm year;
	run;
data firmyr6;
	set firmyr6;
*perfect information returns;
	retg=(priceyrendg-lag(priceyrendg))/lag(priceyrendg);
	retb=(priceyrendb-lag(priceyrendb))/lag(priceyrendb);
*no information returns;
	rete=(priceyrende-lag(priceyrende))/lag(priceyrende);
*scaled earnings;
	snib=nib/lag(priceyrende);
	snig=nig/lag(priceyrende);
	run;
*get "random" good and bad news groups so I have both;
data goodn;
	set firmyr6;
	where ((1<=firm<=250) or (501<=firm<=750));
	news='goodstate';
*learn future truth "each period" reduces to essential one period
	because I am reseting beliefs each period;
	ret=(priceyrendg-lag(priceyrende))/lag(priceyrende);
	sni=snig;
	run;
data badn;
	set firmyr6;
	where ((251<=firm<=500) or (751<=firm<=1000));
	news='badstate';
	ret=(priceyrendb-lag(priceyrende))/lag(priceyrende);
	sni=snib;
	run;
data sim;
	set goodn badn;
	run;
libname research 'L:\faculty\PHDShared\green\research\papers';
data research.sim;
	set sim;
	run;
*/
libname research 'L:\faculty\PHDShared\green\research\papers';
data sim;
	set research.sim;
	run;
proc sort data=sim;
	by firm year;
	run;
data sim;	
	set sim;
	suni=sni-(pprg*snig+(1-pprg)*snib)/lag(priceyrende);
	run;
**********************************************************************

**********************************************************************;
*now I have some dials I can turn:
	 no information ret,full information returns, returns with news during period,good news, bad news,
	 unexpected earnings for good and bad news
	 fc (which imbeds lag for capacity constraints);

*note I may also have to limit negative sales and put a constraint on the returns;
data sim;
	set sim;
	if ret<0 then nret=1; else nret=0;
	if rete<0 then nrete=1; else nrete=0;
	if retg<0 then nretg=1; else nretg=0;
	if retb<0 then nretb=1; else nretb=0;
	if sni<0 then nsni=1; else nsni=0;
	if year=1 or year=10 then delete;
	if id<=5000 then fcd=1; else fcd=0;
	run;
proc means data=sim min p1 p25 median p75 p99 max;
	var ret rete retb retg sni suni;
	run;*version 2 almost all returns are negative because the change in the pv
		is big for any given ni realized, I could add more years...
		so could the second half of basu stuff just be that
		these firms are not expected to survive?;
proc surveyreg data=sim;
	*where fcd=0;
	class year;
	model sni=ret nret nret*ret year/solution;
	run;*so here is the later basu stuff, look at it;
proc surveyreg data=sim;
	class year;
	model suni=ret nret nret*ret year/solution;
	run;
proc surveyreg data=sim;
	*where sni<0;
	class year;
	model sni=ret nret nret*ret fcd fcd*ret fcd*nret fcd*nret*ret year/solution;
	run;


proc surveyreg data=sim;
	class year;
	model ret= suni year/solution;
	run;
proc surveyreg data=sim;
	class year;
	model ret= sni year/solution;
	run;
proc surveyreg data=sim;
	class year;
	model ret= sni suni year/solution;
	run;

proc surveyreg data=sim;
	class year;
	model sni=ret nret nret*ret suni year/solution;
	run;*basu correction?;
proc surveyreg data=sim;
	class year;
	model ret= sni nsni sni*nsni year/solution;
	run;
	*hayn stuff?;
proc surveyreg data=sim;
	class year;
	model ret= sni nsni sni*nsni suni year/solution;
	run;


*change information;
proc surveyreg data=sim;
	class year;
	model rete= sni suni year/solution;
	run;
proc surveyreg data=sim;
	class year;
	model retg= sni suni year/solution;
	run;*why does suni load when ret is perfect information?;
proc surveyreg data=sim;
	class year;
	model retb= sni suni year/solution;
	run;
proc surveyreg data=sim;
	class year;
	model retg= sni suni fcd sni*fcd suni*fcd year/solution;
	run;
proc surveyreg data=sim;
	class year;
	model retb= sni suni fcd sni*fcd suni*fcd year/solution;
	run;
proc surveyreg data=sim;
	class year;
	model ret= sni suni fcd sni*fcd suni*fcd year/solution;
	run;

































proc sql;
	create table firmyr1
	as select *,sum(pvnig) as sumpvnig10,sum(pvnib) as sumpvnib10
	from firmyr3
	group by firm;
	quit;
data firmyr1;
	set firmyr1;
	prc1beg=pprg*sumpvnig10+(1-pprg)*simpvnib10;
	prc1end=pprg*(sumpvnig10-pvnig)+(1-pprg)*(sumpvnib10-pvnib);
	eni=pprg*nig+(1-pprg)*nib;
	run;
data firmyr1;
	set firmyr1;
	unig=nig-eni;
	unib=nib-eni;







data firmyr1;
	set firmyr1;





	eni=((priorprg*nig+(1-priorprg)*nib));
	begprc=priorprg*sumpvnig+(1-priorprg)*sumpvnib;
	endprcg=sumpvnig;
	endprcb=sumpvnib;
	eni=priorprg*nig+(1-priorprg)*nib;
data firmyr;
	set firmyr;
	retg=(endprcg-begprc);
	retb=(endprcb-begprc);
	UNIG=NIG-eni;
	UNIB=NIB-eni;
	run;
proc plot data=firmyr;
	plot retg*unig;
	plot retb*unib;
	plot retg*nig;
	plot retb*nib;
	run;
proc sort data=firmyr;
	by vc;
	run;
proc plot data=firmyr;
	by vc;
	plot retg*unig;
	plot retb*unib;
	plot retg*nig;
	plot retb*nib;
	run;
data firmyrretg;
	set firmyr;
	where year=1 and ((1<=firm<=250) or (501<=firm<=750));
	ret=retg;
	uni=unig;
	ni=nig;
	news='good';
	run;
data firmyrretb;
	set firmyr;
	where year=1 and ((251<=firm<=500) or (751<=firm<=1000));
	ret=retb;
	uni=unib;
	ni=nib;
	news='bad';
	run;
data firmyrret;
	set firmyrretg firmyrretb;
	run;
data firmyrret;
	set firmyrret;
	if ret<0 then d=1; else d=0;
	run;









%mend lagsim;
%lagsim;
options notes;
	if year=1 then vcp=.5 + .1*rannor(0);
	if year=1 then fc=50 + 1*rannor(0);
	run;
*the random makes some sales negative, set them to zero;
data firmyr;
	set firmyr;
	if saleg<0 then saleg=0;
	if saleb<0 then saleb=0;
	run;
options nonotes;
%macro lagsim;
	%do i=1 %to 10000;
data firmyr;
		set firmyr;
	lagsaleg=lag(saleg);
	lagsaleb=lag(saleb);
	lagvcp=lag(vcp);
	lagfc=lag(fc);
	if id=&i and firm=lag(firm) then do;
		saleg=lagsaleg+(lagsaleg*.05)*rannor(0);
		saleb=lagsaleb+(lagsaleb*.05)*rannor(0);
		fc=lagfc;
		lagvcp;
	end;
	run;
	%end;
data firmyr;
	set firmyr;
	drop lagsaleg lagsaleb lagvcp lagfc;
	run;
%mend lagsim;
%lagsim;
options notes;
data firmyr;
	set firmyr;
	if year=1 then priorprg=.5 + .1*rannor(0);
	if id<=5000 then vc=1; else vc=0;
	run;
data firmyr;
	set firmyr;
	if vc=1 then NIG=saleg*(1-vcp);
	if vc=0 then NIG=saleg-fc;
	if vc=1 then NIB=saleb*(1-vcp);
	if vc=0 then NIB=saleb-fc;
	run;
data firmyr;
	set firmyr;
	pvnig=nig/(1+.05)**year;
	pvnib=nib/(1+.05)**year;
	run;
proc sql;
	create table firmyr
	as select *,sum(pvnig) as sumpvnig,sum(pvnib) as sumpvnib
	from firmyr
	group by firm;
	quit;
data firmyr;
	set firmyr;
	eni=((priorprg*nig+(1-priorprg)*nib));
	begprc=priorprg*sumpvnig+(1-priorprg)*sumpvnib;
	endprcg=sumpvnig;
	endprcb=sumpvnib;
	eni=priorprg*nig+(1-priorprg)*nib;
data firmyr;
	set firmyr;
	retg=(endprcg-begprc);
	retb=(endprcb-begprc);
	UNIG=NIG-eni;
	UNIB=NIB-eni;
	run;
proc plot data=firmyr;
	plot retg*unig;
	plot retb*unib;
	plot retg*nig;
	plot retb*nib;
	run;
proc sort data=firmyr;
	by vc;
	run;
proc plot data=firmyr;
	by vc;
	plot retg*unig;
	plot retb*unib;
	plot retg*nig;
	plot retb*nib;
	run;
data firmyrretg;
	set firmyr;
	where year=1 and ((1<=firm<=250) or (501<=firm<=750));
	ret=retg;
	uni=unig;
	ni=nig;
	news='good';
	run;
data firmyrretb;
	set firmyr;
	where year=1 and ((251<=firm<=500) or (751<=firm<=1000));
	ret=retb;
	uni=unib;
	ni=nib;
	news='bad';
	run;
data firmyrret;
	set firmyrretg firmyrretb;
	run;
data firmyrret;
	set firmyrret;
	if ret<0 then d=1; else d=0;
	run;
proc means data=firmyrret mean std median min max;
	class vc;
	var ni uni ret d;
	run;
*this is year 0, only the probability/state changed;
proc surveyreg data=firmyrret;
	model ni=ret d ret*d vc vc*ret vc*d vc*ret*d;
	run;
proc surveyreg data=firmyrret;
	where vc=1;
	model ni=ret d ret*d;
	run;
proc surveyreg data=firmyrret;
	where vc=0;
	model ni=ret d ret*d;
	run;

*now introduce change from realized earnings;
proc sort data=firmyr;
	by firm year;
	run;
data firmyr2;
	set firmyr;
	priorprg=lag(priorprg);
	lagprc=lag(begprc);
	run;
data firmyr2;
	set firmyr2;
	if year=1 then delete;
	run;
data firmyr2;
	set firmyr2;
	drop eni begprc endprcg endprcb sumpvnig sumpvnib retnonews retg retb unig unib;
	run;
proc sql;
	create table firmyr2
	as select *,sum(pvnig) as sumpvnig,sum(pvnib) as sumpvnib
	from firmyr2
	group by firm;
	quit;
data firmyr2;
	set firmyr2;
	eni=((priorprg*nig+(1-priorprg)*nib));
	begprc=priorprg*sumpvnig+(1-priorprg)*sumpvnib;
	endprcg=sumpvnig;
	endprcb=sumpvnib;
	eni=priorprg*nig+(1-priorprg)*nib;
	run;
data firmyr2;
	set firmyr2;
	retnonews=begprc-lagprc;
	retg=(endprcg-lagprc);
	retb=(endprcb-lagprc);
	UNIG=NIG-eni;
	UNIB=NIB-eni;
	run;
data firmyrretg;
	set firmyr2;
	where year=2 and ((1<=firm<=250) or (501<=firm<=750));
	ret=retg;
	uni=unig;
	ni=nig;
	news='good';
	run;
data firmyrretb;
	set firmyr2;
	where year=2 and ((251<=firm<=500) or (751<=firm<=1000));
	ret=retb;
	uni=unib;
	ni=nib;
	news='bad';
	run;
data firmyrret;
	set firmyrretg firmyrretb;
	run;
data firmyrret;
	set firmyrret;
	if ret<0 then d=1; else d=0;
	if retnonews<0 then dnn=1; else dnn=0;
	run;
proc surveyreg data=firmyrret;
	model ni=retnonews dnn dnn*retnonews;
	run;
proc surveyreg data=firmyrret;
	model ni=ret d d*ret vc vc*d vc*d*ret;
	run;
proc surveyreg data=firmyrret;
	model ni=ret d d*ret vc vc*d vc*d*ret dnn dnn*d dnn*ret dnn*d*ret;
	run;
proc surveyreg data=firmyrret;
	model uni=ret d d*ret vc vc*d vc*d*ret;
	run;
proc surveyreg data=firmyrret;
	model ret=ni uni;
	run;
proc means data=firmyrret min max;
	var ret ni uni;
	run;
data firmyrretlimit;
	set firmyrret;
	pctret=ret/lagprc;
	run;
proc means data=firmyrretlimit mean min max;
	var ret ni uni pctret;
	run;
data firmyrretlimit;
	set firmyrretlimit;
	if pctret<-1 then pctret=-1;
	sni=ni/lagprc;
	run;
proc surveyreg data=firmyrretlimit;
	model sni=pctret d d*pctret;
	run;
proc surveyreg data=firmyrretlimit;
	model ni=pctret d d*pctret;
	run;
data firmyrret;
	set firmyrret;
	pctret=ret/lagprc;
	sni=ni/lagprc;
	if pctret<-1 then bigneg=1; else bigneg=0;
	if pctret>4 then bigpos=1; else bigpos=0;
	run;
proc surveyreg data=firmyrret;
	model ni=ret d d*ret bigneg bigneg*d bigneg*d*ret bigneg*ret vc vc*d vc*ret vc*d*ret
			bigpos bigpos*d bigpos*ret bigpos*d*ret;
	run;
proc surveyreg data=firmyrret;
	model ni=ret d d*ret bigneg bigneg*d bigneg*d*ret bigneg*ret vc vc*d vc*ret vc*d*ret
			bigpos bigpos*d bigpos*ret bigpos*d*ret uni;
	run;
proc surveyreg data=firmyrret;
	model uni=ret d d*ret bigneg bigneg*d bigneg*d*ret bigneg*ret vc vc*d vc*ret vc*d*ret
			bigpos bigpos*d bigpos*ret bigpos*d*ret;
	run;
proc surveyreg data=firmyrret;
	where vc=0;
	model uni=ret d d*ret;
	run;
proc surveyreg data=firmyrret;
	where vc=1;
	model uni=ret d d*ret;
	run;





	do i=0 to 1000;
		if firm=i then year=

data type;
	set research.id10000;
	where id<=100;
	firm=id;
	if id<=50 then do;
		FC=50+50*rannor(0);
		CM=1-0;
		HFC=1;
		end;
	else do;
		FC=0;
		CM=1-ranuni(0);
		HFC=0;
		end;
	drop id;
run;
proc sort data=acc;
	by firm;
	run;
proc sort data=type;
	by firm;
	run;
data acc1;
	merge acc(in=a) type;
	by firm;
	if a=1 then output;
	run;
data acc1;
	set acc1;
  	by firm;
  	if first.firm then year = 0;
	year + 1;
run;
options nonotes;
%macro lagsim;
	%do i=2 %to 10000;
	data acc1;
		set acc1;
	newrev=lag(rev);
	if id=&i and firm=lag(firm) then rev=newrev+100*rannor(0);
	run;
	%end;
%mend lagsim;
%lagsim;
options notes;
proc sort data=acc1;
	by id;
	run;
data acc1;
	set acc1;
	do; if rev lt 0 then rev=0;end;
	TotInc=CM*Rev-FC;
	if firm eq lag(firm) then cfo=.9*(Rev*CM)+.1*(lag(rev)*lag(cm))-fc;
		else cfo=.9*(Rev*CM)-fc;
	run;

*assume perfect income smoothing;
proc sql;
	create table acc2 as select
		*,avg(totinc) as smth_ni
		from acc1
		group by firm;
quit;

proc sort data=acc2;
	by descending id;
	run;
data acc2;
	set acc2;
	if smth_ni lt 0 then possmth_ni=totinc;
	else possmth_ni=smth_ni;
	acc=possmth_ni-cfo;
	if firm eq lag(firm) then lead_possmth_ni=lag(possmth_ni); else lead_possmth_ni=.;
	run;
proc sort data=acc2;
	by id;
	run;


proc sort data=acc2;
	by hfc;
	run;
proc corr data=acc2;
	run;
proc corr data=acc2;
	by hfc;
	run;

data acc2;
	set acc2;
	hfc_acc=hfc*acc;
	hfc_cfo=hfc*cfo;
	run;
proc reg data=acc2;
	model lead_possmth_ni=acc cfo hfc hfc_acc hfc_cfo;
	run;

proc reg data=acc2;
	where hfc=1;
	model lead_possmth_ni=acc cfo;
	test acc=cfo;
	run;

proc reg data=acc2;
	where hfc=0;
	model lead_possmth_ni=acc cfo;
	test acc=cfo;
	run;