Most technical analysis studies are concerned with the profitability of technical trading rules and they almost exclusively focus on trend following patterns. In this paper we examine a different kind of technical indicator which suggests a structural relationship between High, Low and Close prices of daily exchange rates.
Since, for a given exchange rate, it can be shown that these prices have different time series properties, it is possible to explore the structural relationships between them using multivariate cointegration methods. This methodology facilitates the construction of dynamic structural econometric models and these are used to derive dynamic out-of-sample forecasts over different time horizons. Compared to standard benchmarks, it turns out that these models have extremely good forecasting properties, even when allowance has been made for transactions costs and risk premia.
Although Technical Analysis1 is often dismissed by academic time series analysts because of its lack of theoretical underpinnings, its popularity has always been high amongst practitioners. For example, in a recent survey of foreign exchange market participants, Menkhoff (1995) showed that there is widespread use of technical analysis by traders and institutional investors for short to medium-term forecasts.
These findings support the results of an earlier survey by Taylor and Allen (1992) of the London foreign exchange market who find that over 90% of dealers use some form of technical analysis in order to derive a short-term forecast. Recent studies also indicate the profitability of technical trading on the basis of simple filter rules (see, for example, Dooley and Shafer (1983), Sweeny (1986) and, more recently, Levich and Thomas (1993)2, where profitability is assessed by outperforming a transactions costs adjusted buy-and-hold strategy3. In this paper we seek to take recent academic work on technical analysis one step further.
In particular, a common feature of extant studies is that the trendfollowing indicators are normally calculated on a close of price basis. However, a different class of technical indicators, widely used among foreign exchange traders, exploits the fact that certain values for a price series appear to have a higher informational content than others. On a daily basis, these correspond to the two extremes: the highest and lowest prices of the day, and the opening and closing price of the market4,5. The difference between High and Low represents the trading range and gives information about the trading activity of a certain period.
Given a specific High and Low, the Close price supposedly contains information about the further price development. So-called Candlestick charts are a graphical way of displaying the different constellations between High, Low, Open and Close prices, where each constellation implies a different forecast (see e.g. Feeny, 1989 )6. While individual candles provide the chartist with information about the trading activity of a certain time period, combinations of consecutive candles form the basis of specific trading signals. Feeny (1989) distinguishes between 24 different types of individual candles and 34 different candlestick formations; however, non-academic sources claim the existence of more than 100 different candlestick formations.
Since it can be shown that High, Low and Close prices of the same exchange rate series have different time series properties, we propose exploring the structural relationships between these prices using multivariate cointegration methods. We find that by restricting the cointegration space it is possible to empirically identify ‘longrun’ relationships in the data that coincide with the underlying structure of this class of technical analysis.
Further, using dynamic modelling techniques we are able to use the identified structural relationships to derive dynamic out-of-sample forecasts over different time horizons. These models produce a creditable out-of-sample forecasting performance in terms of beating a martingale, and also in terms of their ability to generate significant directional ability. Perhaps most pleasingly our forecasting performance does not disappear when risk and transaction costs are allowed for. The outline of the remainder of this paper is as follows.
In the next section we briefly outline the concept of a stochastic, which is a technical indicator that ties down the relationship between High, Low and Close. In section 3 the data set used in this study is discussed and some preliminary statistics presented. The econometric methodology is presented in section 4 along with our estimated results. The forecasting performance of our models is assessed in section 5 in terms of beating a random walk and directional ability, while in Section 6 the implied profitability of the different exchange rate models is compared to a buy-and-hold strategy. A conclusion is presented in Section 7.
1 That is, the use of a broad class of prediction rules for forecasting financial prices.
2 Schulmeister (1987), Leoni (1989) and Menkhoff and Schlumberger (1995) additionally take moving average based indicators and momentums into consideration.
3 However, Cheung and Wong (1997) and Menkhoff and Schlumberger (1995) show that profits derived from technical trading can diminish substantially if a risk premium is accounted for.
4 Since foreign exchange is traded around the world and around the clock, there exists, unlike for stock markets or futures markets, a closing or an opening price. On a daily basis, the Close price therefore corresponds to 5 pm New York time when trading in NY ceases and Sydney prepares to start its currency trading. On weekdays closing prices are identical to the opening prices, since they are recorded at the same point in time, Open and Close prices differ only on weekends and holidays when there is a considerable length of nontrading in the currency market, i.e. that is for weekend: 5 pm NY time on Friday to 8 am Sydney time on Monday.
5 Recently Parkinson (1980) and Garman and Klass (1980) have shown that the efficiency of estimators of price volatility can be increased up to eight times if the classical Close priced based estimators are augmented with information embodied in High and Low prices. An additional motivation for this study lies therefore in the fact that the same might hold for point forecasts.
6 Candlestick charts were already used by Japanese rice traders in the 17th century in order to derive profitable trading rules from a special graphical plot of High, Low, Open and Close prices per time unit, which has a certain similarity to candles.
Prof. Ronald MacDonald, Prof. Norbert Fiess
Next: The Stochastics
Summary: Index