Non-parametric residual bootstrap, extreme empirical distribution.

I am trying to make statistical inference in a stock return predictability study using the bootstrap methodology.

However I am not sure my code works perfectly, since it provides some rather extreme distributions of my t-statistics, (-12 to -24), heres my code:

Anyone who might be able to take a look?

%quarterly_DATA_IS_DP%
%loading data
[dp]=xlsread('Variables.xlsx','5yr','g20:g127');
[nom]=xlsread('Variables.xlsx','5yr','b20:b127');
[real]=xlsread('Variables.xlsx','5yr','c20:c127');
[excess]=xlsread('Variables.xlsx','5yr','d20:d127');

mean_x_t=mean(dp);

%rows and colums in data%
[rows columns]=size(nom);
const=ones(rows,1);

%OLS regression of nom returns on dp, with NW-std.errors.%
results=nwest(nom,dp,5)
results=nwest(real,dp,5)
results=nwest(excess,dp,5)

%(4) in Rapach Wohar
test=ols(excess,const);
e1=test.resid; %residuals from return regression %

lagY=[NaN; dp(1:end-1)];
dp_lag=[const lagY];

%(5) in rapach Wohar)
[b,bint,r]=regress(dp,dp_lag);

%Reisudals from GDP%
e2=r(2:end);
e1=e1(2:end);

%residual vector, with e1 and e2%
e=[e1 e2];

%number of draws (T+100)
T=rows+100;

r_star_t=zeros(T,1);
x_star_t=zeros(T,1);
tstat_v=zeros(200,1);

N=2000; %0000replications
for j=1:N;

%drawing randomly with replacement 
data=datasample(e, T);

%estimating DGP
for i=1:T;
    r_star_t(i)=test.beta+data(i,1);
   if i==1;
    x_star_t(i)=b(1,1)+b(2,1)*mean_x_t+data(i,2);
   else 
       x_star_t(i)=b(1,1)+b(2,1)*x_star_t(i-1)+data(i,2);
   end
end;

%drop 100
r_star_t=(r_star_t(101:end));
x_star_t=(x_star_t(101:end));

%OLS regression of pseudo data
pseudo=nwest(r_star_t,x_star_t,5);
%tstat
tstat_v(j)=pseudo.tstat;

end;

hist(tstat_v)

xlswrite('bootstrap_dp5',tstat_v,'F2:F2002')