به سوی اصول اساسی تجزیه و تحلیل فنی: تجزیه و تحلیل محتوای اطلاعات بالا، پایین و بستن قیمت

Technical analysis assigns a special importance to the Open, High, Low and Close prices in forecasting the mean and volatility of exchange rates. In this paper we propose to investigate the time series properties and the informational content of these different prices, using range and cointegration methods. The application of these methods to a high frequency data set indicates the existence of stable structural relationships and asymmetric information flows, which is supportive of certain predictions of market microstructure models of the foreign exchange market. In sum, we argue that a technical analysis of High, Low and Close prices is a useful way of learning about latent Granger causality in high frequency exchange rates.

Technical analysis (TA) assigns a special importance to the Open, High, Low and Close prices in forecasting the mean and volatility of exchange rates. Candlestick analysis is a popular form of TA that combines Open, High, Low and Close prices for the purpose of charting and forecasting and represents probably the most exhaustive attempt to classify price forecasts according to High–Low–Close constellations.1 Within Candlestick analysis, as well as in other forms of TA, the difference between Open and Close prices serves as a measure of the direction and the extent of intradaily trends. The difference between High and Low prices marks the intradaily trading range and represents a measure of volatility. For many forms of TA, it is the interaction between trend and volatility that is assumed to be informative about future price developments. While TA is often perceived as an example of trading on information unrelated to underlying fundamental rationalisation (Shleifer and Summers, 1990), we argue that this is not the full story: the underlying fundamental rationalisation, at least in part, is provided by the market microstructure of the foreign exchange market. Furthermore, there appears to be a direct link between TA and the favourable statistical properties of the extreme value estimators of Parkinson, 1980 and Garman and Klass, 1980 and Kunitomo (1992).2Blume et al. (1994) have recently shown that TA is valuable when information is costly. In particular, they demonstrate how sequences of volume and prices can be informative and how TA of market data arises as natural components of an agent's learning process. We argue that similar hypotheses can be formulated for the analysis of High and Low prices once we realise that: (1) High and Low prices reveal information about shifts in the demand and supply structure; and (2) changing order flows play an important part in determining market prices. The academic work on support and resistance levels (Curcio and Goodhart, 1992 and DeGrauwe and Decupere, 1992) seems to support the first point. The recent empirical evidence for the latter point is provided by Menkhoff (1998) in a survey of the German foreign exchange market. Menkhoff's results strongly underline the role of order flow analysis in the expectation formation process of foreign exchange market participants. Given that the order flow is unobservable to the uninformed trader, a TA of directly observable High and Low prices allows traders to learn about the underlying market mechanism that drives these order flows. Rather than providing a formal model at this stage, it is the intention of this paper to investigate the informational content of Open, High, Low and Close prices and their value in forecasting volatility, as well as future levels of daily exchange rates. An important feature of our analysis is to realise that High, Low and Close prices, even though derived from the same time series, do not share the same time series characteristics. Although this concept may seem somewhat unusual, the theoretical justification is provided by the time delay concept of Takens (1981). Once we have identified High, Low and Close prices of the same exchange rate as different time series, we then open the possibility of applying a wide range of multivariate modelling techniques to explore the dynamic and structural relationships between High, Low and Close prices. In particular, we are able to show that High, Low and Close prices carry useful information for forecasting the volatility as well as the level of future exchange rates. Since the relevance of High and Low prices in forecasting speculative prices has been recognised by technical analysts for some time, our analysis also raises the question of whether TA (possibly in an intuitive way) exploits latent Granger causality. The remainder of this paper is organised as follows: Section 1 tries to explore the determinants of High, Low and Close prices. Attention is drawn to concepts related to TA, as well as to more general features of the foreign exchange market microstructure. Section 2 investigates the time series properties of High, Low and Close prices for the US dollar bilateral of the German mark (USDDEM), Japanese yen (USDJPY) and British pound (GBPUSD), and provides conclusive evidence in favour of different time series characteristics. Section 3 reveals an information asymmetry between extreme value and return-based volatility measures and shows that a simple High–Low volatility estimator compares favourably to more sophisticated forecasting techniques, including GARCH models. Section 4 reports the results of a multivariate cointegration analysis of High, Low and Close prices and presents evidence from Granger causality tests that point towards the possibility that TA might indeed exploit latent structural relationships in high frequency exchange rate data. The main conclusions are presented in Section 5.

Technical analysis assigns a special role to High, Low and Close prices in forecasting volatility and future levels of exchange rates. The difference between High and Low prices corresponds to the so-called trading range. The trading range is supposed to give information about the trading activity during the course of the day and serves as a simple measure of volatility. Given a specific trading range, the Close price supposedly contains information about the future price development. Different constellations between High, Low, Open and Close prices are considered to imply different forecasts, and Candlestick analysis represents maybe the most comprehensive attempt to classify price forecasts according to High–Low–Close constellations. Many forms of TA try to combine an intradaily trend measure (Open–Close) with an intradaily volatility measure (High–Low) when generating a forecasting signal. This paper represents a first attempt to analyse the informational content of Open, High, Low and Close prices in order to identify a rationale for the use of TA in time series analysis. The theoretical work of Parkinson, 1980 and Garman and Klass, 1980 and Kunitomo (1992) on extreme value estimators, indicates the superior statistical properties of range-based volatility estimators over return-based estimates. Our analysis shows that a range-based volatility measure has a further informational advantage over a return-based volatility measure. In particular, it allows the former to capture important features of the market microstructure of the foreign exchange market, which are related to the information embodied in market turning points. We believe that the superior statistical properties of range-based estimators, together with their ability to exploit asymmetric information flows, provides a reasonable justification for their use in TA. There is, therefore, no reason why range-based estimators should only feature in the forecasting toolbox of a technical analyst. The existence of a stable structural relationship between High, Low and Close prices and the evidence of Granger causality shows that by following a modelling strategy that combines High, Low and Close prices it is indeed possible to generate superior forecasts of volatility as well as future levels of exchange rates. A technical analysis of High, Low and Close prices might therefore present a crude but useful way of exploiting any latent Granger causality which exists in high frequency exchange rates.