Purpose: Devolution of fiscal and administrative autonomy to public schools is a global phenomenon now. Various models of school autonomy have been adopted both in developing and developed countries. The purpose of this paper is to explore the impact of devolution of fiscal autonomy to public primary schools through Parent--Teacher Councils (PTCs) on retention of primary school children in the Khyber Pakhtunkhwa (KP) province, Pakistan. Design/methodology/approach: Two sources of data were used to analyze the research question: Education Management Information System for the years 2006-2011 and 2007-2012, and a specially designed survey questionnaire used to compile information about PTCs from 222 public primary schools in the KP Province. Multiple linear regressions were conducted to examine whether PTC reforms are related to retention rates. An education production function approach was used to examine the effect of "inputs" (PTC reform) on "outputs" (retention). Findings: The regression results indicate that reforms in procedural mechanisms to spend PTC budget and schools with separate classrooms for each grade level are significant in improving retention to the last grade of primary. The results also indicate that retention in all-girls' schools tend to be significantly lower compared to all-boys' schools. Research limitations/implications: First, the integration of data sets resulted in a small sample size, 361 schools, out of which the researcher could visit only 222 schools (10 schools per district) due to time and financial constraints. There may be a probability that with a larger sample size the author findings may look slightly different. However, this is the only current data set collected by the researcher in KP, Pakistan. Second, an ideal way of calculating retention is to track each and every child enrolled in a school over a period of five years and to calculate retention at the end of Grade 5 called true cohort model. However, due to unavailability of such kind of data, a more commonly used method, called reconstructed cohort method, is employed. In this method, data on enrollment by grade are used for six consecutive years, with an assumption that the student flow rates will remain unchanged over time and across grades. Practical implications: The findings of this study provide vital policy input to the Government of Pakistan in particular and other developing countries in general. The study reveals that PTCs have critical impacts on educational outcomes, school productivity and return on public sector educational investment thus providing an impetus for further strengthening of PTC and community participation. Besides, this study offers significant implications as to how school-based management programs will lead to outcomes under resource scarcity in developing countries. Social implications: The paper has implications for the role of school leadership and community participation and for how to engender community involvement in marginalized areas where communities often do not have the time, resources or confidence to participate in their schools. Besides, community participation in parent--teacher meetings means that the school budget is spent transparently and with consensus. Hence, the chances of misuse of funds are minimized to a considerable extent, a dilemma faced by many developing countries. Finally, the collection of PTC-related data regularly especially details about budget allocated, spent and, the unutilized budget may result in better record keeping, which was found lacking during the visit. Originality/value: The uniqueness and originality of this paper can be gauged from the fact that no systematic study exists with regards to the impact of school autonomy on students' retention to the last grade of primary in KP province -- a poor and conflict-ridden region in a low-income country (Pakistan). Also, the data collection from primary and secondary sources was not an easy task. However, the researcher as a civil servant has to use personal contacts to collect primary and secondary data. Hence, this study is unique and first of its kind in nature. No such research has been conducted so far by any researcher, especially in KP.