The first entry in a line of the file mglass.dat is transformed, chaotic 
data generated from a deterministic, discretized, Mackey-Glass system.  The 
transformation and the parameter settings of the system were chosen to make the 
chaotic data resemble differenced log price data from financial markets such as 
the adjusted price data in the files bp.dat or nyse.dat.  The second entry in a 
line is the untransformed data.  

    The discretized Mackey-Glass system is defined by the nonlinear 
autoregression 

   X(t) = X(t-1) + 10.5 * ( 0.2*X(t-5)/(1.0 + X(t-5)**10) - 0.1*X(t-1) ).

From the initial values  

   X(-1) = 1.01052
   X(-2) = 0.99081
   X(-3) = 1.00277
   X(-4) = 1.01380
   X(-5) = 0.97908

the equation was iterated 15010 times.  The last 10010 observations were 
retained.  The retained data were transformed using 

  eps = 1.e-3
  tmp(t) = X(t) - min(X)
  tmp(t) = (1.0 - 2.0*eps)*tmp(t)/max(tmp) + eps
  Y(t) = qt(tmp(t),6)

where qt(.,6) is the quantile function of a 6 degree freedom t distribution.

    In each line of mglass.dat, the first item is Y(t), the second is X(t).  
The data can be read in SAS with 

  data mglass;
    infile "mglass.dat";
    input y x;

and in Splus with

  tmp <- matrix(scan("mglass.dat"),ncol=2,byrow=T)
  y   <- tmp[,1]
  x   <- tmp[,2]