Traffic flow prediction is essential to improve the efficiency of urban transportation. However, traffic flow data exhibits unique spatio-temporal characteristics, which pose significant challenges to traffic forecasting. In this paper, a Spatio-Temporal Lagged Graph Convolution Network (STLagGCN) is proposed to model the multi-scale periodic features and lagged spatial correlations of traffic flow. Considering that traffic historical data contains multi-scale periodic characteristics, the Period-Patch is adopted to divide the time series into subsequence-level patches, and fuse spatio-temporal features extracted from these patches. A Lagged Spatio-Temporal Convolution (LagSTC) block is designed to capture lagged spatial correlations among nodes by Lagged Graph Convolution Network (LagGCN) and temporal features by Temporal Convolution Network (TCN). Experimental results on two real-world traffic flow datasets demonstrate that STLagGCN outperforms state-of-the-art baseline methods in traffic forecasting performance.