Correct way to use Simulate to predict timeseries

I am trying to create an ARIMA model to fit a set of data, that I can then use to predict a future set. I am currently using a data set (attached) that I found online as a test. I am using a smaller sub set of the data as a test fit, then once I can verify the model seems to fit, I will recreate the model on the full data set. Right now, using the Econometrics App i have created a Model that seems to fit very well. I the use

Test = readtable('TestData.xlsx');

Speed = Test(1:49,2);

Speed2 = Test(49:50,2);

Y = simulate(ARMA,50,'Y0',Speed);

plot(Y);

hold on

plot(Speed2);

Where ARMA is the exported model from the Econometrics App, Speed is the set of data used to create the model, and Speed2 is the actual data set I am trying to predict.

This is the created ARMA model for the entire data set

This is the model for Speed.

The models themselves seem to fit well within the app, but if I simulate the next 50 steps from the second model (ARMA(1,1) for Speed) teh resulting plot does not correlate with the actual data.