漸進預算理論和斷續均衡理論是政府預算學界主要的描述性理論。漸進預算理論認為決策者只擁有有限理性，且因要兼顧政治現實，編列預算時往往僅根據前一年度的預算進行微調，斷續均衡理論亦建立於有限理性的基礎上，但欲同時解釋小幅和大幅變動。在台灣，政府編列預算時往往參照往年預算，因此，政府和學者過去較少關注預算變動。本研究分析台灣各縣市2001年至2018年的預算資料。第一部分，我們檢視歲入、歲出、經常支出、資本支出年增率的機率分配圖，討論它們是否符合漸進預算理論或斷續均衡理論的預期。第二部分，我們針對歲入增減、歲入變動、歲出增減和歲出變動等四個變數進行追蹤資料分析，討論造成預算增減或變動的主要原因。本研究發現經常支出較符合漸進預算理論的預期，歲入、歲出及資本支出則較符合斷續均衡理論的預期。其次，本研究的迴歸模型發現選舉相關變數對預算變動的影響不容小覷，舉例而言，選後一年的歲入、歲出增減比例較選舉年低，而選前的歲出變動則較選舉年大。此外，若預算曾出現大幅變動或歲入結構較多元，則較不易再出現大幅變動。我們建議政府應正視預算變動的成因並及早因應可能的預算變動。

Incremental budgeting theory and punctuated equilibrium theory (PET) are two of the predominant descriptive budgeting theories in the past half-century. Incremental budgeting suggests that budget-makers only make incremental changes to the existing budget because of their bounded rationality and political reality. PET also follows the concept of bounded rationality but tries to capture both incremental changes and punctuations in one theory. In Taiwan, it is believed that governments usually make their budgets based on the previous year’s budgets. Thus, government officials and researchers seldom pay attention to budget changes. Here, we analyze the budgets of Taiwanese local governments from 2001 to 2018. First, we discuss the patterns of budget changes in Taiwan. We examine the distribution plots of four types of budgets (total revenue, total expenditure, operating expenditure, and capital expenditure) to see if they follow incremental budgeting theory or PET. Second, we employ panel data analysis and construct four regression models (revenue changes, expenditure changes, and their absolute values) to examine the causes of budget changes. We find that fluctuations in operating expenditure follow the expectations of incremental budgeting theory, while fluctuations in total revenue, total expenditure, and capital expenditure meet the expectations of PET. In our regression models, we find that elections play an important role in affecting the magnitude of budget changes. For example, budget increases in the years after elections are significantly lower than those in election years. The magnitude of budget changes in the year before elections is greater than that in election years. We also find that the history of punctuation and revenue diversification are negatively related to budget changes. This research has significant policy implications, as governments should take those possible causes of budget changes into consideration to devise sound budget decisions and fiscal policies.