Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Business cycles --- Econometric models --- Nonlinear systems --- Econometric models

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

This paper considers the uncertainty associated with upcoming Federal Open Market Committee (FOMC) announcements and the extent to which the market begins to set up for such announcements well before they actually occur. We demonstrate that markets set up well in advance of known announcement days; as a result, there is often less uncertainty in the period immediately preceding an FOMC announcement, despite greater volume of activity, as the market has already incorporated anticipated signals. We consider the relative importance of both macro announcements and central bank officials' speeches and congressional testimony in shaping market expectations. We find substantial evidence of anticipatory effects; these results are particularly relevant as the Fed develops its communication strategy to achieve an orderly exit from its program of quantitative easing.

Zonder onderwerpscode: economie

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

This paper considers the uncertainty associated with upcoming Federal Open Market Committee (FOMC) announcements and the extent to which the market begins to set up for such announcements well before they actually occur. We demonstrate that markets set up well in advance of known announcement days; as a result, there is often less uncertainty in the period immediately preceding an FOMC announcement, despite greater volume of activity, as the market has already incorporated anticipated signals. We consider the relative importance of both macro announcements and central bank officials' speeches and congressional testimony in shaping market expectations. We find substantial evidence of anticipatory effects; these results are particularly relevant as the Fed develops its communication strategy to achieve an orderly exit from its program of quantitative easing.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

This Ph.D. thesis focuses on financial transaction data and volatility. Transaction data capture the characteristics of financial transactions (e.g. transaction time, transaction price, transaction volume, bid and ask price) as they take place on an exchange. Volatility refers to the degree to which asset prices tend to fluctuate. The thesis improves the use of transaction data for market microstructure analysis of the New York Stock Exchange (NYSE). This type of analysis contributes to a better understanding of the functioning of financial markets. Furthermore, the thesis improves the estimation of a standard volatility model and proposes volatility estimators that are much more efficient than classic volatility estimators. Better volatility (model) estimators contribute to better investment decisions and risk assessment within an economy. I start by explaining the recent focus of financial research on transaction data, followed by the focus on volatility, and then present the contributions of each chapter of the thesis to the literature. Databases of transaction data, also called tick data, only became publicly available in the 1990s. Before, financial research and analysis had been mainly based on daily data, i.e. daily averages, closing prices, etc. It soon became clear that this new type of data has specific advantages and offers new research opportunities. For example, higher-frequency data allow more accurate measurement of volatility. The thesis contributes to the growing research on this issue. However, more is not always better. The new data have their own features such as unequally-spaced observations, non-synchronous trading, intra-day seasonal effects, measurement errors due to bid-ask spreads and reporting difficulties, which brought new challenges. Only when these features are satisfactorily dealt with can the advantages of high-frequency data be fully exploited. The thesis also contributes to the literature that seeks solutions to this type of problems. Volatility enters as an essential ingredient in many financial computations, like portfolio optimisation, option pricing and risk assessment. Despite its importance, volatility remains an ambiguous term for which there is no unique, universally accepted definition. The main approaches to compute volatility are by historical indicators computed from daily squared returns, from econometric models such as GARCH, or by indirect computation from option prices based on a pricing model such as Black-Scholes'. Following the introduction of transaction databases, new estimators that exploit intradaily price dynamics have been proposed in the literature. The thesis also presents new estimators along this line. Before the attention switched to measuring volatility, the financial econometrics literature already contained a lot of research on the modelling of volatility. Since the 1980s, starting from the observation of volatility clustering, i.e. periods of high volatility versus low volatility, many models of volatility have been developed that produce and improve forecasts. To this end, Stochastic Volatility (SV) models were developed. These models treat volatility as unobserved, driven by a separate process. Their very nature makes them hard to estimate, however. The thesis points out that a simple estimation method (Generalized Method of Moments, GMM) should be reconsidered to estimate stochastic volatility models. While the chapters of this thesis have a common theme, each chapter can be seen as a separate entity addressing different well-defined issues within financial econometrics. Chapter 1 proposes a new procedure to determine the time of the prevailing quote relative to the time of the trade for New York Stock Exchange data. At the NYSE, trades and quotes are recorded separately, receiving their own time stamp. As a result, trades and quotes are subject to different and varying lags, which makes it hard to reconstruct the sequence of trades and quotes. For market microstructure analysis that is based on trade and quote data at high frequency, it is important to be able to reconstruct this sequence, as mismatching potentially affects the analysis. The procedure put forward in Chapter 1 tests whether the quote revision frequency around a trade is contaminated by quote revisions triggered by a trade, and then determines the smallest timing adjustment needed to eliminate this contamination. An application to various stocks and sample periods shows that the time difference between trade and quote reporting lags varies across stocks and time. The procedure takes this variation into account and hence offers a stock- and time-specific update to the Lee and Ready (1991) 5-second rule, which is often applied in this literature. Chapter 2 contributes to the extensive literature on the estimation of stochastic volatility models. Due to the fact that in SV models the mean and the volatility are driven by separate stochastic processes, volatility is unobservable, which makes SV models hard to estimate. This chapter presents analytical results that may be used to improve and assess the quality of GMM-based estimation of SV models. GMM, while not asymptotically efficient, is still the simplest estimation method for SV models currently available. In particular, we derive closed-form expressions for the optimal weighting matrix for GMM estimation of the SV-model with AR(1) log-volatility, and for the asymptotic covariance matrix of the resulting estimator. The moment conditions considered are generated by the absolute observations, which is the standard in this literature, or by the log-squared observations. We use the expressions to compare the performance of GMM and other estimators that have been proposed, and to optimally select small sets of moment conditions from very large sets. A Monte Carlo study shows large efficiency gains for iterated GMM estimation if it is based on the analytical optimal weighting matrix compared to the case where this matrix is estimated. Chapter 3 proposes new estimators of volatility based on quantiles of the price series, under the assumption that prices are observed without noise. It develops unbiased and consistent estimators of the diffusion coefficient based on quantiles of either the Brownian motion or the Brownian bridge. These estimators are shown to be much more efficient than the range-based estimators of Parkinson (1980) and Kunitomo (1992), where the range is the difference between the supremum and infimum. In particular, efficiency is improved by using more quantiles in the estimation. Moreover, two methods are presented that turn any of the unbiased estimators into consistent estimators. One way to obtain consistency is to apply the unbiased estimators to subintervals and then to average the subinterval estimators. This corresponds to a generalization of the realized range estimator of Christensen and Podolski (2005) and Martens and van Dijk (2007). Furthermore, a new type of consistent estimator based on permuted subintervals is presented. The quantile-based estimators provide an interesting alternative to the existing realized volatility and realized range estimators. Chapter 4 deals with the time-discreteness bias and noise bias of quantile-based volatility estimators when applied to high-frequency data. The former bias is a result of volatility estimators being derived in continuous-time, but applied to discrete-time observations. Despite being derived in continuous time, quantile-based volatility estimators turn out to be fairly robust to the time-discreteness bias except if the estimator is based on price extrema or the number of observations is very small. Analytical and simulation-based bias corrections are presented to deal with the latter cases. Furthermore, attention is given to the bias introduced when the estimators are applied to a price series perturbed by noise. In practice, this noise is due to market microstructure effects, e.g. the transaction price bouncing between bid and ask prices, implying that the ‘true’ price is not observed. A simulation-based noise-bias correction is proposed that deals even with the case in which the noise distribution is unknown. The bias corrections allow the practitioner to exploit the efficiency gain of quantile-based volatility estimation at high frequency. Financiële transactiedata en volatiliteit vormen het onderwerp van dit doctoraat. Transactiedata omvatten de eigenschappen van financiële transacties (bv. de transactietijd, transactieprijs, hoeveelheid, bied- en vraagprijzen, enz.) zoals deze plaatsvinden op een beurs. Volatiliteit verwijst naar de mate waarin prijzen van financiële producten fluctueren. Het doctoraat verbetert het gebruik van transactiedata voor analyse van de micro marktstructuur van de New York Stock Exchange (NYSE). Dit soort analyse draagt bij tot een beter begrip van het functioneren van financiële markten. Verder draagt het doctoraat bij tot het beter schatten van een standaard volatiliteitmodel en stelt het volatiliteitschatters voor die veel efficiënter zijn dan klassieke volatiliteitschatters. Beter volatiliteit(model)schatters dragen bij tot betere investeringsbeslissingen en risicobeheer binnen een economie. Ik leg eerst de recente focus van financieel onderzoek op transactie data uit, gevolgd door een uitleg over de focus op volatiliteit, en bespreek dan de bijdrage van elk hoofdstuk. Databanken met transactiegegevens werden enkel publiek beschikbaar in de jaren 90. Voordien werd financieel onderzoek vooral gebaseerd op dagelijkse gegevens zoals daggemiddelden en sluitingsprijzen. Het werd vlug duidelijk dat deze nieuwe bron aan gegevens voordelen bood en nieuwe onderzoeksmogelijkheden opende. Volatiliteit kan bijvoorbeeld preciezer gemeten worden op basis van dergelijke hoge frequentie data. Het doctoraat draagt bij tot de groeiende literatuur over dit onderwerp. Meer is echter niet altijd beter. De nieuwe gegevens hebben hun eigen kenmerken die nieuwe uitdagingen met zich meebrachten: ongelijke verdeling van observaties over de tijd, niet-synchrone handel, intradag seizoenseffecten, meetfouten wegens het aanbod-vraag ecart, moeilijkheden met rapportering, enz. Enkel wanneer dit soort eigenschappen voldoende in rekening worden genomen, kunnen de voordelen van hoge frequentie data volledig worden benut. Het doctoraat draagt ook bij tot de literatuur die oplossingen zoekt voor dit soort problemen. Volatiliteit vormt een essentieel ingrediënt in velerlei financiële berekeningen zoals portefeuille optimalisering, het prijzen van opties en risicobeheer. Ondanks het belang van volatiliteit blijft het een ambigue term waarvoor geen eenduidige en algemeen aanvaarde definitie bestaat. Volatiliteit wordt vooral berekend op basis van historische maatstaven die gebaseerd zijn op dagelijkse prijsveranderingen, econometrische modellen of indirect afgeleid worden uit optieprijzen. Sinds de introductie van databanken met transactiegegevens werden nieuwe schatters voorgesteld in de literatuur die de intradag prijsdynamiek uitbuiten. Het doctoraat voegt hier nieuwe schatters aan toe. Voordat de aandacht oversloeg naar het meten van volatiliteit bevatte de financieel econometrische literatuur al veel onderzoek over het modelleren van volatiliteit. Beginnend van de observatie van volatiliteitclusters, dat zijn periodes van hoge volatiliteit versus lage volatiliteit, werden sinds de jaren 80 veel volatiliteitmodellen ontwikkeld die projecties voortbrengen en verbeteren. Hoofdstuk 1 behandelt een probleem dat zich voordoet met data van de NYSE, een belangrijke bron van transactiegegevens. Transacties en prijsaanduidingen worden op deze beurs afzonderlijk geregistreerd met elk hun eigen tijdsaanduiding. Hierdoor vertonen transacties en prijsaanduidingen verschillende vertragingen bij hun registratie wat het moeilijk maakt om de volgorde nadien te reconstrueren. Voor analyse van de micro marktstructuur is het echter belangrijk om deze volgorde te kunnen reconstrueren aangezien een verkeerde schikking de analyse kan beïnvloeden. In hoofdstuk 1 wordt een procedure voorgesteld die toelaat om de gepaste tijdsaanpassing te bepalen per aandeel en periode. Hoofdstuk 2 behandelt het schatten van Stochastische Volatiliteitmodellen. Volgens deze modellen wordt volatiliteit niet geobserveerd en wordt het gedreven door een apart proces. Die eigenschap maakt hen moeilijk om te schatten. Hoofdstuk 2 toont aan dat een eenvoudige schattingsmethode (Generalized Method of Moments, GMM) dient te worden herbeschouwd voor het schatten van stochastische volatiliteitmodellen. Analytische resultaten worden gepresenteerd die kunnen worden aangewend om de kwaliteit van schattingen op basis van GMM te verbeteren en te beoordelen. Hoofdstuk 3 introduceert een nieuwe set volatiliteitschatters die gebaseerd zijn op kwantielen van de intradag prijsreeks. Deze schatters vormen een interessant alternatief voor de schatters gebaseerd op intradag prijsveranderingen. De nieuwe schatters halen hun voordeel uit het feit dat kwantielen robuuster zijn voor ruis - dat aanwezig is op hoge frequentie - in vergelijking met intradag prijsveranderingen. Hoofdstuk 4 maakt de schatters klaar voor de praktijk. De nieuwe schatters, die in continue tijd werden afgeleid, worden hier aangepast aan discrete tijd en ruiscorrecties worden voorgesteld.