Portal de Periódicos da CAPES

Inferring trading dynamics for an OTC market: the case of the euro area overnight money market

2010; Taylor & Francis; Volume: 11; Issue: 9 Linguagem: Inglês

10.1080/14697680903515706

1469-7696

Renaud Beaupain, Alain Durré,

Complex Systems and Time Series Analysis

Click to increase image sizeClick to decrease image size Acknowledgements The authors are grateful to Stefano Bragantini, Massimo Buratti, Valentina Carloni and Tommaso Frigerio for providing the data and very useful comments and explanations. The authors also thank workshop participants at the European Central Bank and at the AFFI 2008 Annual Meeting (Lille, France) and three anonymous referees for their valuable remarks and suggestions. The views expressed in this paper are those of the authors exclusively and do not necessarily reflect the views of the European Central Bank or the Eurosystem. Notes †For an empirical investigation of the lending relationships in the overnight money market, see Precup et al. (Citation2005) or Iori et al. (Citation2007). ‡According to the Statute of the European System of Central Banks (ESCB), all credit institutions established in the euro area are subject to the minimum reserve system. These reserves are remunerated over the reserve maintenance period at the average of the marginal rate on the Eurosystem's main refinancing operations (MROs). The reserve requirements are determined by the amount of the corresponding institution's liabilities with a maturity up to two years and exceeding EUR 100,000. For further details, see ECB (Citation2008). §Indeed, the balance sheet data of each institution subject to reserve requirements referring to the end of a given calendar month are used to calculate the reserve base for the maintenance period starting in the calendar month two months later (ECB, Citation2008). †For a detailed description of the features of the Eurosystem's operational framework, see ECB (Citation2008). ‡See ECB (Citation2003, Citation2008) for a description of the refinancing operations of the ECB and the evolution of their features over time. §It is worth mentioning that the overnight transactions seem to have recently gained in importance on the account of the financial distress. According to ECB (Citation2009), overnight transactions would have significantly increased in 2009. ¶It may happen, for instance, that a bank makes a deposit with another bank at the end of the day at a rate below the ECB's rate for the deposit facility. However, being rather rare and, by nature, not representative of standard trading, such observations were excluded from our data (see section 3). They however point to the preferencing relationships that sometimes occur in this segment. ⊥The other main reference in the euro area money market is the euro interbank offered rate (EURIBOR) which is the rate prevailing in the interbank market at maturities from one month to 12 months. See, in particular, ECB (Citation2004) and Durré and Nardelli (Citation2008) for a detailed description. ∥Overnight transactions currently represent more than 85% of e-MID's total turnover. The operation of the platform is detailed in appendix A. †Quotes do not, by definition, offer any guarantee that they will be representative of the realized price of effective trades. This is even more relevant for the information on transactions in the over-the-counter segment reported by wire services. ‡A pure estimate of the duration of quote updates on the electronic platform is not available (given the co-existence of proposals and request for quotes, see appendix A). Nevertheless, it is worth mentioning that a higher proportion of anonymous quotes on Reuters was observed in the early stages of the financial turmoil in August 2007 while the daily number of total quotes appeared to decrease. §Records of transactions on e-MID must have a date, time, price, quantity, and trade direction. Similarly, Reuters quotes must have a date, time, bid price, ask price and a bid–ask spread lower than 25 bps. Trades or quotes with prices falling outside the corridor determined by the deposit and lending rates of the European Central Bank are removed. Finally, we only retained trades or quotes recorded after 08:30:00 and before 18:00:00. ¶When transaction prices deviate from the daily average by more than three standard deviations of the corresponding day, the trade is eliminated. Although various thresholds of standard deviation have been tested (up to five standard deviations), a threshold value of three standard deviations has been chosen given the specificities of the overnight money market to eliminate very small-sized trades, which mostly characterize specific transactions between financial institutions and are not informative for the purpose of our analysis. ⊥Glosten and Milgrom (Citation1985) indeed show that market participants post their bid and ask prices around their personal evaluation of the underlying value. Relying on the quote midpoint as a pre-trade approximation of the EONIA is particularly relevant when using the best quotes available, that is, the highest bid and the lowest ask prices from the market. †As a robustness check, we additionally examine the implied effective spread introduced by Roll (Citation1984). This alternative measure of transaction costs on the platform yields very similar conclusions. The results are available upon request. ‡The first measure was aimed at better segmenting maintenance periods through non-overlapping operations, whereas the second and third were intended to eliminate the impact that any expectations of changes in the key ECB rates might have on counterparties' bidding behaviour in MROs. See, in particular, ECB (Citation2003, Citation2005) for more details. §Given the strong persistence of our measures, we alternatively examine the spread between our measures and the prevailing minimum bid rate. This analysis confirms the robustness of our conclusions. The results are available upon request. †To ascertain the robustness of our results, we apply the following loss functions: mean squared errors (MSE), mean squared proportional errors (MSPE), mean absolute errors (MAE), mean absolute proportional errors (MAPE), adjusted mean absolute proportional errors (AMAPE), QLIKE and LINEX. ‡Our Diebold–Mariano test statistic is built as DM = LOSS SWTP − LOSS Y where LOSS is one of the loss functions presented above and Y is EWMQ (panel A), EWTP (panel B), SWTP.b (panel C) or SWTP.s (panel D). †As the authors show, the prior belief provided by the mid-quote of the prevailing bid and ask prices is significantly less informative than the posterior belief formed from effective transaction prices. †The null hypothesis of mean equality is examined by means of ANOVA F-statistics. We additionally compute Wilcoxon–Mann–Whitney, Median Chi-Square, Kruskal–Wallis and van der Waerden test statistics to check the equality of medians between the EONIA and its proxy. Variance equality is finally investigated in a series of Siegel–Tukey, Bartlett, Levene and Brown-Forsythe test statistics.