In today's world, massive on-demand mobility requests are dispatched every hour on ride-sourcing platforms, however, customers later cancel quite a number of these confirmed orders. This paper makes the first attempt to look into this customer confirmed-order cancellation behaviour based on a two-month, hourly average dataset of Didi Express in Shanghai provided by Didi Chuxing. The mean ride distance and pick-up distance of cancelled orders are observed to be obviously longer than those of completed orders of the same time period, reflecting an obvious impact of travel cost on customer decisions of order cancellation. However, the correlation between the mean customer confirmed-order cancellation rate (COCR) and the mean customer waiting time for pick-up of cancelled orders is significantly negative. Shanghai, like many other cities around the world, has coupled ride-sourcing and taxi markets, and as such, this counter-intuitive phenomenon can be explained as an outcome of the lower (higher) chance of meeting vacant taxis while waiting for pick-up during peak (non-peak) hours. We formulate COCR as a function of customer waiting time for ride-sourcing vehicles and cruising taxis, the penalty strategy by the platform for cancellation of confirmed orders, and customers’ own characteristics (i.e., ride distance, value of time, perceived psychological cost of order cancellation, additional safety concern over ride-sourcing), and propose a system of nonlinear equations to depict the complex interactions between the ride-sourcing and taxi markets considering the probability of ride-sourcing cancellation after order confirmation. With the proposed model, we replicate the observed lower COCR under a higher demand rate and longer waiting time for pick-up through numerical examples, and highlight the potential improvement of platform profit that can be achieved by appropriately designed penalty/compensation strategies.