We show that the standard Granger causality test for assessing informational efficiency between financial markets is misspecified in the presence of market-microstructure noise, a pervasive feature of financial data. Although the test remains statistically valid, its economic interpretation is flawed: predictability from microstructure noise is misread as information flow. We propose a new test robust to such noise and apply it to credit markets, overturning established results. The corporate bond market, not the CDS market, leads in price discovery; there is no evidence of insider trading in CDS; and bond transactions contain more timely information than quotes.