Too many significant ACFs

Hi guys, got a fairly simple issue here. I am attempting to produce an ACF and PACF in order to get a suitable ARMA model. The data had a trend, however I differenced it. 35 of my ACFs are significant, while 3 of my PACFs are significant, giving me an ARMA (35,3). The ARIMA functions gives me (2,0,1). Clearly, there is a major contradiction. To me something seems wrong. Can you tell me if this is possible/likely and if not, could you suggest any of the main areas I may have gone wrong. Feel free to ask any questions to get any more info. Thanks for reading.

Significant ACFs don't necessarily correspond to the number of AR parameters. Think of an AR(1) with parameter 0.9. The ACF will be 1, .9, .81, .72,...

Hi, thanks for the response. I see, so with many significant ACFs (e.g. 35), how else would I decide the AR number?

You might go with the number of significant PACFs, or you might accept what the ARIMA function says.

If the ACF values decline slowly toward zero while the PACF values drop quickly after the first three, you have an autoregressive process of order three, or AR(3), not MA(3). The fact that ARIMA chose AR(2) makes sense. After accounting for that, ARIMA found a MA(1) process, which would be hard to see in the correlograms.