Haplotype phasing is essential for identifying disease-causing variants with phase-dependent interactions as well as for the coalescent-based inference of demographic history. One of approaches for estimating haplotypes is to use phase-informative reads, which span multiple heterozygous variant positions. Although the quality of estimated variants is crucial in haplotype phasing, accurate variant calling is still challenging due to errors on sequencing and read mapping. Since some of such errors can be corrected by considering haplotype phasing, simultaneous estimation of variants and haplotypes is important. Thus, we propose a statistically unified approach for variant calling and haplotype phasing named HapMonster, where haplotype phasing information is used for improving the accuracy of variant calling and the improved variant calls are used for more accurate haplotype phasing. From the comparison with other existing methods on simulation and real sequencing data, we confirm the effectiveness of HapMonster in both variant calling and haplotype phasing.