I have an OLS model with macro-economic variables like GDP, Unemployment rate as my independent variables. While testing for serial correlation up to order 4 with Breusch–Godfrey test (using proc autoreg in SAS), the test is passing for AR(1) i.e. no serial correlation of order one is present. However the test fails for higher orders.
What is the implication of such a test result? Can i go ahead and fit an OLS model instead of error correction models since the residuals don't have an AR(1) structure or will it induce some model risk?