آیا تجزیه و تحلیل فنی روزانه در بازار سهام ایالات متحده ارزشمند بوده است؟

This paper investigates whether intraday technical analysis is profitable in the U.S. equity market. Surveys of market participants indicate that they place more emphasis on technical analysis (and less on fundamental analysis) the shorter the time horizon; however, the technical analysis literature to date has focused on long-term technical trading rules. We find, using two bootstrap methodologies, that none of the 7846 popular technical trading rules we test are profitable after data snooping bias is taken into account. There is no evidence that the market is inefficient over this time horizon.

The use of past price movements to predict future price movements (technical analysis) has been popular with the investment community for a considerable period of time. When the key word “technical analysis” is typed into the Internet search engine Google, 201,000,000 urls are located compared to only 71,300,000 urls for “fundamental analysis”.2 Despite this widespread acceptance and adoption by practitioners, academics have traditionally treated technical analysis with disdain. It has been described by Malkiel (1981) as an “anathema to the academic world” due to its conflict with market efficiency, one of the central pillars of academic finance. Surveys of market participants and journalists consistently find that these individuals place more emphasis on technical analysis (and less emphasis on fundamental analysis) the shorter the forecasting horizon (e.g., Carter and Van Auken, 1990, Allen and Taylor, 1992, Lui and Mole, 1998 and Oberlechner, 2001). More specifically, respondents place approximately twice as much weight on technical analysis for intraday horizons as they do for one-year horizons. Despite market participants ascribing the most value to short-term technical analysis, the academic literature has focused on testing the profitability of long-term technical trading rules.3 Most studies find that technical analysis is not profitable once transaction costs are taken into account (e.g., Allen and Karjalainen, 1999, Bessembinder and Chan, 1998 and Olson, 2004). However, Corrado and Lee (1992) and Lee, Chan, Faff, and Kalev (2003) point out that technical analysis may still have merit as a value-adding “overlay” strategy to assist investors such as fund managers in better timing the buying or selling of stocks as part of their normal trading activities. Under this scenario the stock trades would have occurred in the normal course of business so the transaction costs are already factored in. This paper considers the value of equity market technical analysis on an intraday basis using 5-minute Standard and Poor's Depository Receipts (SPDR) data. In doing so, several contributions are made. First, to the best of our knowledge, this is the first paper to consider the profitability of intraday equity market technical analysis. This is important because it is heavily used by practitioners over this time horizon, and recent papers by Kavajecz and Odders-White (2004) and Osler (2003) find evidence of order clustering that is consistent with the propositions of technical analysis. Given that the price pressure from order clustering is a short-term phenomenon, it seems reasonable to expect this to lend more support to intraday technical analysis than daily or monthly technical analysis. In addition, the short-term nature of intraday technical analysis also means that any profitability is extremely unlikely to be driven by time varying risk premia.4 Second, the use of actual transactions data for the Standard and Poor's Depository Receipts (SPDRs), the exchange traded fund that replicates the S&P 500 index, by this paper has several advantages. Previous studies, such as Neely and Weller (2003) and Osler (2000), analyze the value of technical analysis on intraday foreign exchange market data, but the absence of foreign exchange market trade data necessitates that these papers estimate transaction prices based on bid and ask quotes. In addition, the choice of SPDR data has several advantages over the index data that has been used by the majority of longer-term technical analysis papers. Indices are not tradable in their own right so any technical trading signals would therefore be unable to be implemented without purchasing each of the index components in the correct proportions. Moreover, as Day and Wang (2002) document, tests of technical trading rules on index data can be biased due to non-synchronous trading. Finally, technical analysts claim that technical analysis is most reliable on actively traded stocks (Morris, 1995). By the end of 1999 there was $19.8 billion invested in SPDRs, and in 1998 the daily dollar volume was the highest of any stock (Elton et al., 2002). We purposely study the January 1, 2002 to December 31, 2003 period to give us an insight into any difference between trading rule performance in bull and bear markets.5 The S&P 500 declined 21.2% in 2002 and increased by 21.9% in 2003. Third, the choice of 7846 trading rule specifications from five rule families (Filter Rules, Moving Average Rules, Support and Resistance Rules, Channel Breakouts, and On Balance Volume Rules), which were widely publicized prior to the start of this study, allows a fair test of market efficiency. Miller (1990) points out that the development of financial theories alters behavior so testing models with data from before the models were developed is less than adequate. Finally, unlike the previous intraday technical analysis literature (on the foreign exchange market), which does not conduct robust statistical tests of the significance of profits they document, we apply a suite of tests. These are the Brock, Lakonishok and LeBaron (1992) (hereafter BLL) approach of fitting null models to the data, generating random series and comparing the results from running the rules on the original series to those from running on the randomly generated bootstrapped series, and the so-called White's Reality Check bootstrapping technique (Sullivan, Timmerman, and White (1999), hereafter STW) which adjusts for data snooping bias. To the best of our knowledge, this is the first paper to utilize both these techniques. Our results clearly demonstrate that intraday technical analysis is not profitable on the SPDR series over the 2002–2003 period. While there is evidence that a small number of rules are profitable prior to any formal adjustment for data snooping bias, none of the five rule families produced a statistically significant profitable rule once this was taken into account. Given these results, we conclude that there is no evidence of consistent inefficiency in the intraday SPDR data. The remainder of the paper is organized as follows: Section 2 outlines the technical trading rules used to test market efficiency and the steps that were taken to minimize data snooping bias. Data and methodology are presented in Section 3. Section 4 contains the results and Section 5 concludes the paper. 2. Technical trading rules employed It is clear that the application of new trading rules, or new specifications of existing trading rules, to historical data introduces the chance of data snooping bias. It is quite possible that the rules have been tailored to the data series in question and are only profitable because of this. If this is the case, there is nothing to suggest that the rules will be profitable out of sample, or that someone would have chosen those exact specifications ex ante to form a profitable trading rule. To prevent data snooping bias, Pesaran and Timmerman (1995, p. 102) conclude that “as far as possible, rules for predicting stock returns should be formulated and estimated without the benefit of hindsight.” Both Lo and MacKinlay (1990) and Lakonishok and Smidt (1988) maintain that this new data approach is the best protection against data snooping. A second approach to minimizing data snooping bias involves adjusting the statistical significance of a particular trading rule by taking account of the universe of all trading rules from which it is drawn (e.g., STW, 1999). While it is not possible to quantify the entire universe of trading rules that one rule might have been chosen from (e.g., Ready, 2002), the inclusion of a wide range of different rules does significantly reduce the risk that any given rule's profitability is due to chance. To minimize the chance of data snooping, we follow both approaches.6 More specifically, we use the five broad types of rules that received wide publicity prior to the start of our data. These have been published in many different papers and were succinctly summarized in STW (1999).7 Like STW, we also apply a data snooping adjustment technique. Filter Rules are the first rule family we consider. Standard filter rules, which were first introduced by Alexander (1961), involve buying (short-selling) after price increases (decreases) by x% and selling (buying) when it decreases (increases) by x% from a subsequent high (low). We define subsequent highs and lows in two ways. The first definition is the highest (lowest) closing price achieved while holding a particular long (short) position. The second is the most recent closing price that is less (greater) than the e previous closing prices. We also investigate filter rules that permit a neutral position. This is accomplished by closing a long (short) position when price decreases (increases) y% percent from the previous high (low). We also consider holding a long or short position for a prespecified number of periods, c, effectively ignoring all other signals generated during this time. The second rule family we consider are Moving Averages, which appear to have been developed by Gartley (1930). In their most basic form, a buy (sell) signal is generated when the price moves above (below) the longer moving average, because at this point a trend is considered to be initiated. We also investigate rules that generate a buy (sell) signal when a short moving average (e.g., 10 periods) moves above (below) a longer moving average (e.g., 200 periods). We investigate the impact of applying two filters. Firstly, we require the shorter moving average (or price) to exceed the longer moving average by a fixed multiplicative amount, b. Secondly, we require a buy or sell signal to remain valid for a prespecified number of periods, d, before action it taken. As with the filter rules, we also consider holding a position for a prespecified number of periods, c. Support and Resistance or “Trading Range Break” rules were developed by Wyckoff (1910). In their most simple form, these involve buying (short-selling) when the closing price rises above (falls below) the maximum (minimum) price over the previous n periods. The extreme price level that triggers a buy or a sell can also be defined as the most recent closing price that is greater (less than) the e previous closing price. As with moving average rules, we also impose a fixed percentage band filter, b, and a time delay filter, d. Again, positions can be held for prespecified number of periods, c. A fourth family of rules, Channel Breakouts, are similar to Support and Resistance Rules. A channel is said to occur when the high over the previous n periods is within x% of the low over the previous n periods, not including the current price. Following STW (1999), the channel breakout rules we test involve buying (selling) when the closing price moves above (below) the channel. Long and short positions are held for a fixed number of periods, c. Additionally, a fixed band, b, can be applied to the channel as a filter. On-Balance Volume (OBV) Averages are the final rule family we consider. This indicator, which was popularized by Granville (1963), is calculated by keeping a running total of the indicator each period and adding (subtracting) the entire amount of daily volume when the closing price increases (decreases). We then apply a moving average of n periods to the OBV indicator, as per STW (1999). The OBV trading rules employed are the same as for the moving average rules, except the variable of interest is OBV rather than price.

We investigate whether intraday technical analysis is profitable in the U.S. equity market. Surveys of market participants consistently find that technical analysis is more highly valued the shorter the time horizon, but prior work has focused on the profitability of long-term technical analysis. Studying intraday equity market technical analysis has several other aspects. Recent work has found evidence of order clustering which may provide theoretical support for technical analysis. Since this is a short-term phenomenon the impact of order clustering is likely to be more pronounced on intraday technical analysis. We examine the profitability of 7846 rules from five major rule families (Filter, Moving Average, Support and Resistance, Channel Breakout, and On-Balance Volume Rules). These rules were well documented in the literature prior to the start of our sample so we propose that data snooping bias is minimised. As a further precaution against this, we formally adjust our test statistics to account for this. Our results clearly demonstrate that intraday technical analysis is not profitable on the SPDR series over the 2002–2003 period. While the evidence that a small number of rules are profitable prior to any formal adjustment for data snooping bias, none of the five rule families produced a statistically significant profitable rule once this was taken into account. Given these results, we conclude that there is no evidence of consistent inefficiency in the intraday SPDR data.