Why lag1, may be 2 or 3, of the PACF plot seem to be more significant than a higher lag? I am attempting to interpret the PACF plot generated from the daily return of the market index.
As it can be seen on the graph, AR(1) seems to be the most appropriate model since the partial autocorrelation is significant at lag 1. However, according to a lot of examples shown in the internet, I found that most of the example of the PACF plot interpretation mainly concentrates on the lag1, or may be 2 or 3, regardless of the significant lag at higher lag. For instance, even though that the partial correlation is significant at 23, 30 and 31, with a lag length of 31, but if the lag 1 is significant then it is going to be AR(1). Or if lag 1 is insignificant, then the series is generated on a random manner. So, [ 1] I would like to ask that why we mainly rely on lag1 
[2]According to my PACF plot, How should I make a comment on the lag 23,30 and 31. Since, for instance, the return at t and t-31 are compared at lag 31 and it is significant, so does it mean it is possible to predict the return on a monthly basis?

https://1drv.ms/x/s!Al4Zyp7bfeKdxkez0UG-Q-kTgKIM: Series of return data, excel
Many thank
Thomas
 A: ARIMA modelling can be a very powerful tool BUT like all model specifications there are assumptions. An ARIMA model is not a panacea.  Good econometricians are taught early to test "model specification assumptions" . A critical assumption is that if you have daily data then you must not omit any observations. You data has missing values(dates eg. 4.21.12 where the financial markets were closed.) My observation/finding could only be found by examining the time series in question.
Other assumptions of ARIMA modelling is that there needs to be a constant variance in the (to be found) error process . Also there needs to be no pulses/level shifts/seasonal pulses ( like a monday effect) or local time trends. You data clearly has "spikes' which need to be explained either via dummies or known events.
Often times in daily data there are deterministic effects like day-of-the-week , month-of-the-year and even week-of-the-month or day-of-the-month. These need to be factored into the equation culminatimg in a Transfer Function with possible ARIMA structure. 
An example of a daily model can be found here http://autobox.com/cms/index.php/afs-university/intro-to-forecasting/doc_download/53-capabilities-presentation starting at slide 48
When an OP asks a very general question like this one it is insufficient to deliver a "Sermon from the Mount" like read the book ! but rather to provide possible specific help regarding logic flaws in their approach. Actual assumption testing much like an earlier Thomas can often be very useful.
