加载中...
Marching Against the Wind: Typhoon Disasters and Banks' Risky Behaviors
气候变化已成为全人类面临的重大挑战,对金融经济体系产生深远影响。台风作为高频、高强度发生的极端天气事件,不仅对经济活动造成巨大破坏,也对金融部门产生强烈冲击,进而引发一系列连锁反应与反馈行为。本文以2010—2019年中国295家地方商业银行为研究对象,匹配3452家A股上市公司的经营数据,结合Holland风场模型计算城市级台风破坏力指数,实证考察台风灾害对银行风险行为的影响,并检验企业全要素生产率与企业固定资产损失的中介效应,进一步分析由此引发的银行风险偏好变化与信贷决策等系列连锁反应与反馈行为。本文实证分析发现:第一,台风灾害显著提高了银行不良贷款率,增加了银行信用风险,且沿海地区银行不良贷款率的上升幅度显著高于内陆地区;第二,企业固定资产损失上升与企业全要素生产率水平下降是台风灾害影响银行不良贷款率的重要渠道;第三,进一步分析显示,台风灾害对金融机构影响的传导过程并非单一,台风灾害先冲击实体经济,再传导至金融体系,最后又反馈至实体经济。具体而言,地区受台风灾害冲击后,企业生产经营陷入“经营困境”,该困境传导至银行层面,致使银行不良贷款率上升,即被动风险承担增加;这进一步影响银行未来的信贷决策,如收紧信贷规模、降低风险偏好,即主动风险承担降低;这又反馈至企业层面,提高企业的融资成本,最终放大了台风灾害对整个经济金融活动的影响。本文为银行气候风险治理提供了新的经验证据。
Abstract
Climate change has become a major challenge to all of humankind with profound impacts on the financial and economic system. As an extreme weather event occurring with high frequency and intensity, typhoons not only cause great damage to economic activities but also have a powerful impact on the financial sector, which in turn triggers a series of chain reactions and feedback behaviors. This paper empirically examines the impact of typhoon disasters on banks' risk behavior by taking 295 local commercial banks in China during 2010-2019 as research objects, matching the operational data of 3452 A-share listed companies and combining these with the Holland wind field model to calculate a typhoon destructive power index at the city level. The paper also examines the mediating effects of corporate total factor productivity and corporate fixed asset losses and it further analyzes the resulting series of chain reactions and feedback behaviors including changes in banks' risk appetite and credit decisions. The empirical analyses yield several findings. First, typhoon disasters significantly raise banks' non-performing loan ratios (NPLRs) and increase banks' credit risk, and they raise banks' NPLRs significantly more in coastal areas than in inland areas. Second, the rise in corporate fixed asset losses and the decline in corporate total factor productivity levels are important channels through which typhoon disasters affect banks' NPL ratios. Third, further analyses show that the process by which typhoon disasters' effect on financial institutions is transmitted is not monolithic. Typhoon disasters first hit the real economy, then are transmitted to the financial system, and finally swing back to the real economy. Specifically, after being affected by a typhoon disaster, the production and operation of enterprises fall into "economic difficulties", which are transmitted to the bank level, making banks' NPLR increase (i.e., an increase in passive risk-taking). This further affects banks' future credit decision-making, such as causing them to tighten the extent of credit and reduce their appetite for risk (i.e., a decrease in active risk-taking). This in turn feeds back to the enterprise level, increasing the cost of financing for enterprises and ultimately magnifying the impact of the typhoon disaster on overall economic and financial activities. This study provides new empirical evidence for banks' climate risk governance.
台风灾害显著提高银行不良贷款率
台风破坏力指数每增加一个标准差,银行不良贷款率平均上升约0.36个百分点。这一效应在沿海地区银行中更为显著,沿海地区银行的NPLR上升幅度显著高于内陆地区。该结论在替换解释变量、PSM-DID、工具变量法等多种稳健性检验下依然成立。
企业资产损失是台风影响银行风险的重要渠道
台风灾害导致企业固定资产损失增加21.6%,企业固定资产损失每增加1%通过中介效应传导至银行,使银行不良贷款率显著上升。表明企业实物资本受损是台风灾害影响银行信贷风险的重要机制。
企业全要素生产率下降是另一重要传导路径
台风灾害显著降低企业全要素生产率水平(系数为-0.6628),企业生产效率的下降进一步增加了银行的不良贷款率,形成从实体经济到金融体系的风险传导链条。
台风灾害引发银行风险偏好下降与信贷收紧
面对台风灾害造成的信贷风险上升,银行会降低主动风险承担,具体表现为收紧信贷规模(系数:-5.1532),而这会反馈至企业层面,提高企业融资成本,最终放大台风灾害的总体经济影响。
核心解释变量
Typhoon:城市-年份层面的台风破坏力指数,基于Holland风场模型计算,考虑台风风速、路径、距离等因素构建连续型指标。
被解释变量
NPLR:银行不良贷款率,衡量银行信用风险(被动风险承担)。进一步分析中使用银行风险偏好或信贷(主动风险承担)、企业融资成本等变量。
样本与数据
以2010-2019年中国295家地方商业银行为研究样本,匹配3452家A股上市公司运营数据。剔除关键变量缺失样本后,银行层面样本量为2950个城市-银行-年份观测值,企业层面包含28263个企业-年份观测值。银行财务数据来自银行年报,企业数据来自A股上市公司披露。
识别方法 / 模型设定
利用城市层面连续型台风破坏力指数的差异作为外生冲击。首先采用OLS固定效应模型,控制城市和年份固定效应,考察台风对银行NPLR的总体影响;进一步通过PSM-DID(以是否受台风严重冲击分组)、工具变量法(以台风历史登陆频率作为工具变量)、替换核心变量定义等方法进行因果识别。
内生性及稳健性检验
稳健性检验包括:①使用PSM方法构造处理组与控制组,并在此基础上进行DID估计(系数为0.0040);②使用工具变量法处理潜在内生性(以2014年台风登陆路径作为IV,第一阶段F值为148.25);③使用不同年份窗口的DID分析排除预期效应;④替换台风破坏力指数计算方式(如500km、1000km范围);⑤排除极端样本影响等。