Standard deviation of OFHEO HPI "returns" increases linearly with sampling period length

I've been playing with OFHEO House Prise Index data this week. It's open access data and can be downloaded into CSV file from OFHEO's web site.

I copied Los Angeles MSA data into this spreadsheet. Columns A-G are original data columns. Columns C and are the year and the quarter. Column is E is HPI.

Columns G-I are HPI "returns" with quarterly, semi-annual and annual sampling. "Return" is r(t) = HPI(t) / HPI(t-lag) - 1, where lag is 1 for quarterly, 2 - for semi-annual and 4 - for annual. At the bottom of the spreadsheet there's standard deviation of these returns: 0.0253, 0.0482 and 0.0969. As you can see, it increases linearly with sampling period length.