Rare-variant aggregation tests are often used to increase the power to detect rare-variant associations, and can be performed using publicly available GWAS summary statistics. Accurate estimates of single-variant test statistic covariances are needed, but are often unavailable due to data size and inability to share individual-level genetic data. In this study we extend a previously proposed method of estimating covariance from an external reference panel to rare-variant aggregation tests. We apply our method to UK Biobank exome sequence data and observe squared Pearson correlation coefficients comparing aggregation test -log10(p-values) generated using individual-level data and UK Biobank external reference panels between 0.984 and 0.996 at a reference panel of 1,000 across multiple simulation settings. By using an African American reference panel (n=1,000) from the InPSYght study, we also show that our approach is robust to misspecification of the reference panel ancestry.