KHO THƯ VIỆN 🔎

Developing a volume forecasting model

➤  Gửi thông báo lỗi    ⚠️ Báo cáo tài liệu vi phạm

Loại tài liệu:     PDF
Số trang:         40 Trang
Tài liệu:           ✅  ĐÃ ĐƯỢC PHÊ DUYỆT
 













Nội dung chi tiết: Developing a volume forecasting model

Developing a volume forecasting model

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelolume Forecasting ModelBogdan Batrinca1. Christian w, Hesse2 and Philip c. I releaven’AbstractThis study builds a series of models to predict trading

volume in European markets using different statistical methods. The analysis considers a number of aspects, such as special events (e.g. MSCT rebalanc Developing a volume forecasting model

es, futures expiries, or cross-market holidays), day-of-thc-week effects, and the volume-priee relation asymmetry, in order to perform contextual one-

Developing a volume forecasting model

step ahead prediction. We investigate the prediction error for each calendar circumstance to infer a cross-stock event-oriented switching model for vo

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelelding the best performance for the most recent observations.JET. classification numbers: C32, C52, C53. Ct 14. G1.5Keywords: Trading volume, expiry d

ay effect, holiday effect, behavioral finance. European stock market, feature selection1 IntroductionMeasuring trading performance is a challenging re Developing a volume forecasting model

search area, but there are certain factors that have a clear influence on the overall trading performance, such as the market impact, which is the eff

Developing a volume forecasting model

ect caused by a market participant who buys or sells shares, consisting in the extent to which the price goes upward for a buy order or downward for a

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelreated [IJ. Market impact can move the prices adversely, leading to decreased profits or turning profitable strategics into losing strategics.’Departm

ent of Computer Science. University College London. Gower Street. London well’ 6BT, UK.Department of Computer Science. University College London. Gowe Developing a volume forecasting model

r Street. London WCiE 6BT.UK.‘Department of Computer Science. University College London. Gower Street. London WC1E 6BT. UK.Article Info: Received. Sep

Developing a volume forecasting model

tember 1. 2016 Revised: September 29. 2016.Published online: January 5. 20172Bogdan Batrinca el al.The execution style of an order drives the extent o

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelinto smaller orders over a longer period in order to trade slowly with a low market impact. Therefore, predicting the trading volume as a measure of l

iquidity is of vital importance to forecast the expected market impact.The aim of this study is to propose a switching volume prediction model by fitt Developing a volume forecasting model

ing a variety of models that employ different machine learning methods and considering endogenous and exogenous variables that may potentially impact

Developing a volume forecasting model

the trading volume. This is motivated by the importance of optimally sizing an order for minimising the market impact and ultimately improving the tra

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelicipalc by creating opportunity cost and price uncertainly. Therefore, predicting the trading volume helps better determine the degree of participatio

n in the market.The primary’ focus of this study is to fine-tune the models and identify’ the optimal model given the market context at a certain poin Developing a volume forecasting model

t in lime, in order to achieve optimal prediction accuracy and model stability. We arc investigating the error breakdown by different model types and

Developing a volume forecasting model

day’s that matter (e.g. holidays, expiries, days-of-the-week etc.).Each stock exhibits different levels of trends, volatility’, and magnitude in their

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelredictive models for seven machine learning techniques: ordinary leasl squares, stepwise regression (i.c. ordinary least squares with sequential featu

re selection), ridge regression, lasso regression, fc-nearest neighbours with arithmetic average, ^-nearest neighbours with inverse distance weighting Developing a volume forecasting model

, and support vector regression. For each statistical method, we iterate every’ stock in our pan-Luropcan slock universe consisting of 2.353 slocks, e

Developing a volume forecasting model

very training window type (i.c. mov ing/sliding vs. growing) and every window size (i.e. 1-month. 3-monlh. 6-monlh. 1-year. 2-ycar windows). We also I

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modely are ultimately’ used to make stock-specific predictions. We tit these models in isolation and aim to determine a performance metric for each method

and window type. Eventually, we shift from a sialic process to an adaptive process and construct a switching dynamic model, which switches between the Developing a volume forecasting model

se models based on the current context (e.g. regular trading day. cross-market holiday, futures expiry’, MSCĨ rebalance, certain day-of-the-week etc.)

Developing a volume forecasting model

. The proposed model is a virtually switching model as it does not switch per sc. We arc post-processing the model performance and investigate the per

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelamodel, which is a stock-specific out-of-sample model that selects the best initial stock-specific model on a I -month and a 3-month rolling window ba

sis, depending on the recent performance of the initial stock-specific models that arc trained independently of each other.The rest of the study is st Developing a volume forecasting model

ructured as follows: section 2 reviews the key findings that led to our model choice in this study (e.g. the volume-price relation asymmetry’, the day

Developing a volume forecasting model

-of-the-week effect, the expiry day effect, and the cross-market holidays effect) and outlines the methods employed in this analysis; the market and c

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting model this computationally expensive analysis, followed by a methodological introduction of the cross-stock models and the stock-specific models;Developing

ã Volume Forecasting Modelthis is followed by section 5, which presents the main findings of this study, including a performance breakdown of the mod Developing a volume forecasting model

els, and introduces the switching model and the out-of-sample stock-specific metamodel: eventually, section 6 provides a conclusion of this analysis a

Developing a volume forecasting model

nd discusses the obtained results.2 BackgroundPrevious studies provided empirical evidence for the volume-price relation and its asymmetry, and the ex

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelndings are summarised below and arc followed by a review of the statistical methods employed in this analysis.2.1Volume-Price Relation and AsymmetryTh

e price-volume relation is of great importance for this study as most of the behavioural literature focuses on the impact of certain anomalies on pric Developing a volume forecasting model

e returns, while trading volume is the main focus of this study. Price changes represent the market response to new information, whereas the trading v

Developing a volume forecasting model

olume indicates the level of information disagreement among investors [21. Although the literature on a potential relation between price changes and v

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting model et al. |5J provided empirical evidence that trading volume is correlated with historical price indicators (i.c. intraday range and intraday return fo

r the previous day. and overnight return for the previous night, which acts as a proxy for the opening auction volume, i.e. more recent information) a Developing a volume forecasting model

nd that volume exhibits autoregression, where we employed lagged time series volume data (i.e. raw past observations) and also smoothed lagged time se

Developing a volume forecasting model

ries (i.e. moving average of past observations, which acts as a low-pass filter effect in the data). The formulae for the intraday return, intraday ra

Journal of Applied Finance & Banking, vol. 7, no. Ị, 2017,1-40 ISSN: 1792-6580 (print version), 1792-6599 (online) Scienpress Lid. 2017Developing a Vo

Developing a volume forecasting modelj is the previous trading day, whose price and volume information is available;P’10.se intraday return log ratio: PintradayRtn = logoffintraday range

log ratio:PintTadayRnR — lnn Ijiow1 . P:Tn overnight return log ratio: PovemightRtn =-1-2-3In general, there are two key representations of the volume Developing a volume forecasting model

-price relation, where trading volume is positive correlated either with the magnitude (i.e. absolute value) of the price change [6], i.e. |Ap|, or wi

Developing a volume forecasting model

th the price change per se (i.e. the raw value of the price change), i.e. Ap [7] [8]. The asymmetric relation in the latter representation exhibits a

Gọi ngay
Chat zalo
Facebook