One of the major challenges involved in risk aggregation is the lack of risk data. Recently, researchers have found that mapping financial statements into risk types is a satisfactory way to resolve the problem of data shortage and inconsistency. Nevertheless, ignoring off-balance sheet (OBS) items has so far been regarded as the usual practice in risk aggregation, which may lead to deviations in conclusions. Hence, we improve the financial statements based risk aggregation framework by mapping OBS items into risk types. Based on 487 quarterly financial statements from all 16 listed Chinese commercial banks over the period 2007–۲۰۱۴, we empirically study whether the overall impact of OBS activities and the individual impact of each of the OBS risk types on total risk depend on bank size. Moreover, this research divides the sample into two subsets, during and after the subprime crisis, to find out how the subprime crisis affects risks of Chinese banks. Our empirical results show that although OBS credit risk is positively linked to total risk while OBS operational risk is negatively linked to total risk for both large and small banks, the overall impact of OBS activities on total risk depends on bank size. The overall OBS activities are positively related to the large bank’s total risk while they are negatively related to the small bank’s total risk. Besides, we also found that it is the increase of liquidity risk and market risk that leads to the larger total risk of Chinese banks during the subprime crisis.

Introduction

Some characteristics of off-balance sheet (OBS) activities, such as blind expansion and high risk, made the existence of OBS activities a key factor that caused destabilization during the subprime crisis (Brunnermeier 2009). Basel II, however, was widely seen as having failed to adequately capture the risks posed by OBS activities (Acharya and Richardson 2009; Blundell-Wignall and Atkinson 2010). Essentially, OBS risk should be regarded as an indispensable part of a bank’s overall risk because both on- and off-balance sheet activities create bank risks (BCBS 1986). Basel Committee has already made great strides in strengthening regulatory capital framework to cover risks, whatever the source (BCBS 2010). Thus, a reliable risk aggregation model to capture both on- and off-balance sheet risks is urgently needed. Broadly, risk aggregation refers to a quantitative risk measurement method that incorporates multiple types of risk (Li et al. 2015). One major challenge in risk aggregation is the risk data used for establishing marginal risk distributions (BCBS 2003). Many previous studies have attempted to use simulated risk data to measure credit risk, market risk and liquidity risk (Dimakos and Aas 2004; Acerbi and Scandolo 2008), which can hardly replace the real data. For the operational risk, external real data are often used to supplement insufficient internal loss data. However, some remain skeptical of the external operational risk data (BCBS 2003; Chavez-Demoulin et al. 2006). Thus, the shortage and inconsistency of risk data limit the reliability and validity of risk aggregation results. Recent research has, instead, used publicly available industry-wide data from a set of commercial banks’ financial statements to develop empirical proxies for different risk types. Although financial statements data have some drawbacks, such as lower reporting frequency (usually published quarterly), different accounting standards across the world (Bae et al. 2008) and poor accounting quality (Saito 2012), collecting risk data from financial statements is still a satisfactory way to resolve the problems of data shortage and data inconsistency. Some have attempted to aggregate marginal risks based on-balance sheet data. Kretzschmar et al. (2010) implement a fully-integrated risk analysis based on-balance sheet asset positions. However, the exclusion of OBS derivatives from asset portfolios weakens the effectiveness of qualitative conclusions. Given the importance of OBS items, Drehmann et al. (2010) not only take account of balance sheet assets and liabilities, which have been considered by Alessandri and Drehmann (2010) for integrating credit and interest rate risk, but also pay attention to OBS items. Such a modification makes the hypothetical bank reflect a real commercial bank more accurately. Mapping profit and loss (P&L) items from income statement into risk types is another feasible way to obtain risk data. As researchers have realized that risk is defined in terms of earnings volatility (Rajan 2006), P&L items from income statement that are created by earnings volatility can be used as proxies for risks (Kuritzkes and Schuermann 2007). Thus, Kuritzkes and Schuermann (2007) get risk P&L successfully by mapping income statement items of US banks into risk types. Given the significant accounting difference between income statements in US and China, Li et al. (2012) use data of risk P&L to measure Chinese banks’ risks by establishing a mapping relationship between Chinese banks’ income statements and risk types.