Capturing depositors’ expectations with Google data
Falko Fecht ()[1], Stefan Thum ()[2], Patrick Weber ()2
Keywords: Depositor expectations, Google, Deposit insurance, Competition for depositors, Bank runs, Interest rate sensitivity of non-financial agents
1. Introduction
Recent regulatory initiatives following the European Banking Union introduce a bail-in of bank debt. While potentially fostering market discipline and reducing tax payers' burden in bank resolutions, bail-ins might aggravate depositor panics. During the recent crisis many banking systems in Europe experienced episodes that resembled traditional bank runs. The failure of Northern Rock, the Icelandic banking crisis, collapse of the Cypriot banking system as well as the deposit freeze in Greece are frequently cited examples showing how doubts about the willingness and/or ability of governments to insure retail depositors arguably lead to widespread panics and the demise of deposit taking financial institutions. However, the extent to which self-fulfilling panics indeed aggravated the crises and whether an extensive and credible insurance could stop those panics is heavily debated.
In this paper we contribute to the question whether government backstops are essential in mitigating depositors' panics. We exploit particularities of the German banking system that provide an excellent setup to test for the role of governmental deposit guarantees in investors' withdrawal decisions. German savings banks are typically jointly owned by the municipality and the regional association of savings banks which means that they are de facto government guaranteed. Savings banks compete at the local level mostly with credit cooperatives, which are only backed by the association of cooperative banks. At the wake of the Irish banking crisis also German depositors became anxious about the stability of their bank. This induced the government to introduce in October 2008 a blanket guarantee for all German banks' liabilities. Hence this set-up allows using both the cross-sectional and the intertemporal difference in depositors' insurance to assess the drivers of investors' decision to shift funds from cooperative to savings banks and vice versa. To exploit heterogeneity in depositors' fear of a bank run we use the frequency of Google searches for deposit insurance and related strings at local level. This permits us to assess whether differences in the deposit shift between local savings and cooperative banks are indeed related to local investors' concerns about deposit guarantees.
2. Methods
2.1. Data and Variables
For our empirical analysis, we match three key data sources: First, we obtain data from Google Trends from July 2007 to February 2016. Second, we match this data with the outstanding volumes of overnight deposits from the Bundesbank's Monetary Financial Institutions Balance Sheet Items (BSI) statistics. Corresponding interest rates are obtained from the Bundesbank's Monetary Financial Institutions Interest Rate (MIR) statistics. Both BSI and MIR statistics are monthly statistics with data available at the bank level. For our main regressions, we group deposit volumes and volume-weighted interest rates for savings banks and credit cooperatives at the federal state level.[3]
The key variable of interest is the relative search interest of economic agents for terms related to e.g. deposit insurance or how safe is my money to proxy the fear of the general public about their deposits. Google Trends data show the search interest for a given search term relative to the total Google search volume over time.[4] Since we are interested in whether an increase in the fear of depositors regarding their deposits leads to an outflow of deposits from private (credit cooperatives) to public banks (savings banks), we construct the variable Deposit Shiftj,t which measures total overnight deposit volume at savings banks relative to the total overnight deposit volume at credit cooperatives:
Deposit Shiftj,t=Volume Savings Bankj,tVolume Cooperative Bankj,t
where j is the respective federal state and t is the respective month. For the empirical regressions, we work with the six months difference time series of Deposit Shiftj,t
∆Deposit Shiftj,t=Deposit Shiftj,t-Deposit Shiftj,t-6 .
To control that a deposit shift from private banks to public banks may happen because private banks decreased their overnight interest rate offered to depositors, or because public banks increased overnight interest rates paid on deposits, we construct the variable Interest Marginj,t which measures the spread between the interest rate paid to depositors in a respective federal state by savings banks relative to that paid by cooperative banks
Interest Marginj,t=Interest Rate Savings Banksj,t-Interest Rate Coop. Banksj,t .
Similar to the Deposit Shift variable, we work in our empirical regressions with the six months difference time series, lagged by one month of Interest Marginj,t to account for a potential endogeniety bias
∆Interest Marginj,t=Interest Marginj,t-1-Interest Marginj,t-7 .
2.2. Methodology
For our main empirical models, we estimate a simple standard panel model with robust standard errors of the form
∆DepositShiftj,t=αj+αt+β1∆InterestMarginj,t+β2Googleg,j,t+uj,t
where αj is specified as the j's federal state fixed effect, αt as a monthly time fixed effect and g is the gth Google search term related to deposit insurance. The search term deposit insurance is the only one with a sufficient number of search requests at federal state level. We use simple OLS regression for all alternative search terms that are only available at the level of Germany and include yearly, instead of monthly, time fixed effects.
In order to estimate whether economic agents are sensitive to increases or decreases in the interest rate spread between private and public banks, we separately estimate their reaction to changes in this measure before and after a level playing field between private and public banks has been established (because of the announcement of a public guarantee for all banks in October 2008). In addition to that we add dummy variables that we interact with Google search term(s) to check whether the sensitivity of depositors' fears after the introduction of the guarantee vanished.
∆DepositShiftj,t=αj+αt+β1Guaranteet+β2Googlef,j,t+β3Googlef,j,t*Guaranteet+β4∆InterestMarginj,t-1*Guaranteet+β5∆InterestMarginj,t-1*NoGuaranteet+uj,t
where NoGuarantee is a binary indicator variable equalling one before October 2008 and zero afterwards and Guarantee is a binary indicator equalling one after October 2008 and zero before.
3. Results
Firstly, we can show that Google searches Granger cause both the interest spread and deposit shifts. The results hold independent of the ordering and are robust to alternative Google search terms.
Figure 1. Main regression result (Perspective: State level)
Our analysis provides the following key results. First, local Google searches for "deposit insurance" are highly related to local shifts of deposits from credit cooperative to savings banks. Google searches seem to capture local investors' worries about the safety of their deposits very well suggesting that searches might serve also as an early warning indicator for banking panics. Second, a high interest rate paid by local savings banks relative to local cooperative banks is associated with an increase in the local market share of savings banks, but only in the period after the blanket guarantees introduced a level playing field. Before the blanket guarantees there was no significant effect of rate spreads on the market share of cooperatives relative to savings banks. We apply various robustness checks in the data accumulation as well as in the model setup (fixed effects/ random effects models, use of household/ corporation deposit volumes only, different winsorizing and conversion techniques and different time frames) that do not hamper our results.
4. Conclusions
The results of our analysis allow to draw several conclusions. First and maybe foremost they show that Google searches can indeed be used as a measure for the concern of depositors about the stability of the local banks indicating run-type phenomena in local deposit markets, adding to the fast growing literature following [1]. This proxy permits us to disentangle fundamental factors from effects resulting from depositors' sentiments. Furthermore, related to [2], our results show that concerns about banks' stability matter for the reallocation of funds in a crisis and that even within the banking sector differences in the governmental guarantees matter for deposit flows. However, our results also suggest that governmental support extended during the crisis is not necessarily a timely game changer. Finally, our results indicate that blanket guarantees provided during the crisis led to a level playing field between private and public banks making deposit flows more sensitive to interest rate spreads contributing to the vast literature analysing the distorting effects of public banks on competition and stability in the banking sector.[5] Consequently, insurance extended to all depositors introduced fiercer competition in the deposit market along with potentially more excessive risk-taking.[6]
References
[1] Z. Da, J. Engelberg and P. Gao, The Sum of All FEARS: Investor Sentiment and Asset Prices, Review of Financial Studies 28(1) (2014), 1-32.
[2] V. V. Acharya and N. Mora, A crisis of banks as liquidity providers, Journal of Finance 70(1) (2015), 1-43.
[3] R. Gropp, H. Hakenes and I. Schnabel, Competition, Risk-shifting and Public Bail-out Policies, Review of Financial Studies 24(6) (2010), 2084-2120.
[4] R. Gropp, A. Guettler and V. Saadi, Public Bank Guarantees and Allocative Efficiency, IWH Discussion Papers No 7 (2015).
[5] C. Matutes and X. Vives, Competition for deposits, fragility, and insurance, Journal of Financial Intermediation 5(5) (1996), 184-216.
[6] C. Matutes and X. Vives, Imperfect competition, risk taking, and regulation in banking, European Economic Review 44(1) (2000), 1-34.
4
[1] Frankfurt School of Finance and Management, Chair for Financial Economics
[2] Deutsche Bundesbank, Statistics Department
[3] Both Google Trends and MIR data at the federal state level are constrained to six German states (Baden-Württemberg, Bayern, Hessen, Niedersachsen, Nordrhein-Westfalen, Rheinland-Pfalz). These six states account for approximately 83% of the relevant overnight deposit volumes.
[4] Since these time series feature extreme values around the date of the Merkel/ Steinbrück public commitment in October 2008, we winsorize the Google Trends series.
[5] See, for instance [3] and [4].
[6] This line of argument would, for instance, be supported by the theoretical considerations of [5] and [6].