Can R^2 Metric be used for rolling forecasts?

Question

I forecasted some pricing data and evaluated it using a few of the standard metrics (MSE,MAE,R^2). The forecasting was done using fbprophet and a rolling ARIMA forecast. The graph of the forecast is shown here:

We can see that the rolling ARIMA follows the test data really well (maybe too well? Overfitting?) whereas prophet's forecasts are slightly off.

However, when I calculated the metrics for both these forecasts I get these values:

ARIMA Rolling RMSE:  2.7175816638468993
ARIMA Rolling MSE:  7.385250099676881
ARIMA Rolling MAE:  1.4263079222919917
ARIMA Rolling R2 Score:  0.6559313910482427


Prophet RMSE:  3.349429559727403
Prophet MSE:  11.218678375575704
Prophet MAE:  2.238762881014073
Prophet R2 Score:  0.6490628017042303

Prophet's R2 score seems reasonable it feels, possibly a little low. However the rolling ARIMA forecasts R2 score is only 0.5% higher when I would have expected it to be +90% since it follows the test data so well.

Is there an explanation behind this?

Thanks very much in advance for your answers.