We analyze timing noise from five years of Arecibo and Green Bank observations of the seventeen millisecond pulsars of the North-American Nanohertz Observatory for Gravitational Waves (NANOGrav) pulsar timing array. The weighted autocovariance of the
timing residuals was computed for each pulsar and compared against two possible models for the underlying noise process. The first model includes red noise and predicts the autocovariance to be a decaying exponential as a function of time lag. The second model is Gaussian white noise whose autocovariance would be a delta function. We also perform a ``nearest-neighbor correlation analysis. We find that the exponential process does not accurately describe the data. Two pulsars, J1643-1224 and J1910+1256, exhibit weak red noise, but the rest are well described as white noise. The overall lack of evidence for red noise implies that sensitivity to a (red) gravitational wave background signal is limited by statistical rather than systematic uncertainty. In all pulsars, the ratio of non-white noise to white noise is low, so that we can increase the cadence or integration times of our observations and still expect the root-mean-square of timing residual averages to decrease by the square-root of observation time, which is key to improving the sensitivity of the pulsar timing array.
