Is it okay to use AIC to determine the different arima model from different calculation of data?

So I have used the monthly DJI index adjusted price as my data, and I built up three arima models from the data using log returns, simple returns and the adjusted price respectively.

I knew that generally, the lower the aic, the better the model, but can this apply to my arima model when they came from the same data, but with different calculations?

That shouldn't be a problem with the same models, except, possibly, the adjusted price (depending on how it is transformed from returns).