Forecasting quantity of newborns to better allocate cram school/daycare centers and their resources

Project Details


Fall 2014


Di-Chien Lo, Guerman Shion, Jieng-Wuen Cai





Cram Schools (CS) are specialized schools that train their students to meet particular goals, most commonly to pass the entrance examinations of high schools or universities. People in Taiwan are aware about the importance of early childhood education and development. Also, Day Care Center (DCC) is an ongoing service during specific periods, such as the parents' time at work. Recently large numbers of women workers are entering or returning to the workforce. The growing economy is encouraging parents to spend more on these child development services.
Taiwan has the lowest birth rate in the world, it’s a serious national security threat and definitely a big demographic problem. It’s also a big shock for cram schools and day care centers, facing the plummeting birth rate, they need to figure out solutions to use their resources more efficiently.

The profitability of a CS/DCC is largely determined by the quantity of students. For all providers, the largest expense is labor. Hence, it can be argued that a sustainable competitive advantage can be achieved, if market size of new potential students could be accurately forecasted, it could allow a better management, and ultimately lead an increase in performance. Our basic business objective is to develop a more targeted strategy of resources allocation of CS/DCC in Taiwan, staff (full-time and par-time teachers) and site (size and location).
In this project we are trying to accurately forecast two horizons for the demand of newborns, short-term and long-term, with respective strategies 1 year for staffing purpose and 3 years for allocating branches. By leveraging the power of forecasting and predict the demand of newborns as to “when” and “where” should be opened or closed a branch we aim to obtain an optimized equilibrium of demand and supply.

Some main features to note within the data are: chosen series have two different sizes, collected between 1981-2013 for newborn babies and crude married rate (33 periods, yearly) and 1999-2013 for bank deposit rate, female work rate and female education. (15 periods, yearly). The data shows a twelve year seasonality based on Chinese Zodiac and a downward trend overall. This forecast has ongoing capability, since the data is updated yearly its implementation don’t require automation. In this report, we used data from the Ministry of the Interior of Taiwan.

In the first step we forecast using naïve as a benchmark to compare other models. Test results in model based forecasts proved that the influence of married rate series are statistically significant when predicting the newborn series. Among the data-driven models not one outperformed the benchmark.

The store can increase its profits by preventing overcapacity or under-capacity. Different models work best and are more reliable depending on the horizon, for short-term forecast (one year) we used linear regression and for long-term (3 years) we used naïve as they showed better performance among other models. One should also keep in mind the fact that external data will be available only after a certain lag of at least a year. Based on our research, we recommend a model using a combination of two forecasting techniques that are simple, yet robust and flexible enough to create a reliable prediction.

Application Area: