Available online at www.sciencedirect.

com

Physica A 331 (2004) 639 650

www.elsevier.com/locate/physa

A momentum trading approach to technical analysis of Dow Jones industrials

Hui Wanga; b ; , Ras B. Pandeya
of Physics and Astronomy, University of Southern Mississippi, Hattiesburg, MS 39406-5046, USA b Department of Physics, University of Electronic Science and Technology of China, Chengdu 610054, China
a Department

Abstract A momentum trading approach is presented to examine the Dow Jones industrial components for a period of about past 10 years (19922002). An analogy between the classical dynamics in physics and the stock trade dynamics is used with the momentum, P = mv, where the velocity (v) is a relative price change in a period ( ) and the inertial mass (m) is a normalized trade volume. Extrema in the momentum time series, i.e., the singularities in the driving force provide the signals for executing trades, minima with negative momentum to buy and maxima with positive momentum to sell. Trades are implemented using a momentum threshold (Pc ). A range of periodic cycles ( =5240 days) in time series and trading momentum thresholds (|Pc | =0:01 0.5) are considered and returns (maximum, minimum, accumulative, and average) are examined in detail on the historical DJI data for about a decade (19922002). Frequency of trade is generally higher with smaller periods with the high probability of higher returns at |Pc | = 0:02 0.1 for nearly all stocks in DJI. c 2003 Elsevier B.V. All rights reserved.

1. Introduction Technical analysis of moving averages and momentum based on the assumption of an e cient market is an important part of trading tools in major stock markets [16]. The hypothesis of pure random stock market is common in econometry [7] while it is not considered as explicitly in empirical nance. It is well known that the e ciency of the market varies and econophysics is an attempt to develop a better understanding
Corresponding author. Department of Physics and Astronomy, University of Southern Mississippi, Hattiesburg, MS 39406-5046, USA.

0378-4371/$ - see front matter c 2003 Elsevier B.V. All rights reserved. doi:10.1016/j.physa.2003.08.037

640

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

of econo-dynamics [814]. Numerous variables a ect the econo-dynamics and are reected into the price index of the stock market. These variables include interest rates, sales signals for goods, consumer condence, decits, surplus, emergence of emerging tools and perceived potentials and breakthroughs, the evolving socio-economic dynamics, etc. [15], and probability of unexpected disasters ( ood, earthquakes, accidents, etc.). A variety of practices are exercised in current stock market [13] with an enormous range of technical details some of which appear ad hoc and based on pure visual judgment of graphs and charts on the part of the analysts. Availability of data, statistics, and charts have led to a rather very fast dissemination of opinions with a long list of terminology for various indicators [1]. A scientic analysis, however, requires a systematic observation of trends and relates it to basic laws, fundamental or empirical. In a series of papers, Ausloos and Ivanova [1618] have recently investigated the stock price, active volume, moving averages, and presented an elegant analogy with the fundamental quantities of mechanical physics. For example, they extended the denition of the classical momentum, the average price change during the moving average period to a more technical footing by introducing the generalized momentum. Momentum is dened as P = mv, where mass (m) is the normalized volume transaction and velocity (v), the average rate of price change during the moving average period. Using the stock price of IBM for a decade (1990 2000), they pointed out sign of appropriate signals for the transactions. The analogy between the stock dynamics and the classical mechanics of particles dynamics is an appealing concept to pursue. We would like to follow the same terminology with somewhat di erent denitions for mass, velocity, and momentum. A momentum trading approach is presented to analyze a sector represented by the Dow Jones industrial (DJI) stocks which may help understanding signals for selecting specic stocks from the DJI group. Method and denitions are presented in the next section followed by detailed analysis and conclusions.

2. Method DJI consists of 30 blue chip companies (see Table 1) each with its specic price variations (time series). Returns (prot and loss) of all companies are analyzed for 10 years (19922002) if trades were executed during this period using the following momentum trading approach at appropriate time. In order to examine the rate of price variation, one has to select the unit of time or the period ( ) in which the price change occurs. The period may vary, i.e., hours, days, months, etc., depending on short to long time (t ) analysis. The average velocity w, the rate of price (p) change in unit time, i.e., over the period ( ) is generally dened as w = p(t + ) p(t ) ; (1)

where can vary from short (weeks) to long (months) periods. The rate of price change (w) varies from one stock to another in an index group. In order to reduce the

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650 Table 1 DJI components with their maximum volumes in 10 years (19922002) Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Ticker symbol AA GE JNJ MSFT AXP GM JPM PG BA HD KO SBC C HON MCD T CAT HPQ MMM UTX DD IBM MO WMT DIS INTC MRK XOM EK IP Name ALCOA INC GENERAL ELEC CO JOHNSON& JOHNSON MICROSOFT CP AMER EXPRESS CO GENERAL MOTORS JP MORGAN CHASE PROCTER & GAMBLE BOEING CO HOME DEPOT INC COCA COLA CO SBC COMMS CITIGROUP HONEYWELL INTL MCDONALDS CORP AT& T CORP CATERPILLAR INC HEWLETT-PACKARD 3M COMPANY UNITED TECH CP DU PONT CO INTL BUS MACHINE ALTRIA GROUP WAL-MART STORES WALT DISNEY CO INTEL CORP MERCK & CO EXXON MOBIL EASTMAN KODAK INTL PAPER CO

641

Max-volume 17019200 85888900 50702400 1.57E+08 31975400 33585400 46819400 68643600 36727200 66652300 18691000 28986200 1.22E+08 86215600 35334900 72589600 8575700 55881500 14558000 19894200 21612800 69447300 66870200 24770500 61067600 3.09E+08 32864100 39491200 18155900 10794600

variability and focus on the fractional (or percentage) change, it is convenient to look at the relative price change u(t ) = w(t )=p(t ) : (2)

For the analysis of the entire DJI index, it is useful to consider the relative strength over a specied time ( ) range. A relative velocity v(t ) is dened according to v(t ) = u(t )=uh (t ) ; (3)

where uh (t ) is the highest value of u(t ) among all DJI components during the unit time element ( ). Fig. 1 shows the time series of uh (t ) for = 10 and 60 days. This provides an estimate of the dominant e ect of individual stock in the price variation of the whole index. Larger period seems to have higher values of uh (t ) as expected.

642

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

Fig. 1. Time series of the maximum velocity.

The inertia, i.e., mass (m) associated with trading each stock (i) is dened from its trading volume (V (t )) as m(t ) = V (t )=Vm ; (4) where Vm is the maximum volume of that stock (i) during the entire period (1992 2002). Thus the trading momentum (P (t )), P (t ) = m(t )v(t ) : (5) It is worth pointing out again that the denition of relative velocity (Eq. (3)) and time dependent mass (Eq. (4)) is di erent from those of Ausloos and Ivanova [1618] and Van Der Hart and Slagter [19]. We focus on the time series of momentum (Eq. (5)) for each DJI component. Extrema (minima and maxima) provide signals for trades (buy and sell). Di erent trading thresholds for the momentum time series are analyzed for a period of about 10 years (19922002). Results are provided for the corresponding returns if these trades were executed without including the cost of execution (i.e., commission for trade). Detailed technical analysis of these DJI stocks follows next. 3. Analysis The time series for momentum (Eq. (5)) and price change for each stock in DJI index are examined in detail. Fig. 2 shows a typical variation for the GE and MSFT for only a 3 years period for clarity. Since the momentum is a product of two normalized quantities (price change and volume, see Eq. (5)), its variation with time is constrained with positive and negative values. It shows better patterns than the price change alone which may have wider swings. One may easily nd peaks (maxima) and valleys (minima) in momentum series in a somewhat oscillatory fashion. The magnitude of these extrema and their temporal positions provide signal for trades. For example, one may buy stocks at or near a minimum (valley with negative momentum) and sell it at or near the next maximum (positive momentum) of the momentum series (see Fig. 2). Such executions are valid with the time series of the stock price itself, i.e., buy low and sell high

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

643

Fig. 2. Time series for price and momentum with indicators to buy (upward arrows with negative momentum) and sell (downward arrows with positive momentum) for GE: Pc = 0:1 (a) and for MFST: Pc = 0:3 (b).

strategy is nearly fool proof. However, it is not easy to identify maxima and minima and their upcoming signals so readily in variation of stock price. We believe that the momentum is a better indicator, as Ausloos and Ivanova [1618], for such technical analysis. Because of restrained oscillatory patterns, it is easier to identify peaks and valleys with positive and negative momentum, respectively. Since these extrema are more frequent in momentum than that in the price of a stock itself, there are more opportunities for trades with such momentum trading.

644

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

Fig. 3. Typical trades with buy and sell prices for the Microsoft with |Pc | = 0:01 during 19921999.

Catching these extrema in a rapidly changing market place with nearly instantaneous execution is a very di cult task. Therefore, a momentum threshold (Pc ) needs to be established. A close examination of momentum series during 19922002 of all 30 stocks in DJI index seems to suggest that Pc = 0:05 and 0:10 (of the wide range of |Pc | = 0:01 to 0.5 considered here) could be an appropriate choice for executing trades with = 5240 days. When the momentum falls below the negative value of Pc buy signals are executed while if the momentum runs over the positive Pc on the way up, sell signals are issued. These executions are based on the past historical data for 10 years (19922002) to see how this momentum trading strategy would have worked. For example, in 60 days (3 months) cycle of momentum series for GE, the opportunities to buy stocks (lower points) and to sell (upper points) with all threshold |Pc | 6 0:1 will lead to prot if the commissions for executing the trades are a small fraction of the prot (see Fig. 2). Note that at the selling points (the maxima with the positive momentum), the stock prices are not necessarily at their maximum peak (Fig. 2). In general, the extrema in momentum time series do not necessarily coincide with the corresponding extreme in the price time series. As pointed out by Ausloos and Ivanova [1618], trend in momentum leads the stock price, and therefore, is a good indicator for the trade. Since our denition of momentum is di erent, we cannot make such assertion. However, the momentum is a better indicator and inclusion of the threshold for trade makes it a convenient tool for analysis. Overall the momentum trading with this method shows that the probability of return is relatively high (see below) based on the analysis of DJI index data for about a decade (19922002). As mentioned above, the momentum trades are analyzed for the time series data with the price change during various time intervals ( =5240 days) for di erent momentum thresholds on the past historical data (19922002). Maximum, minimum, accumulated and average returns per trade are examined in detail. Typical trades with buy and sell prices for the Microsoft with |Pc | = 0:01 is shown in Fig. 3 for period 19921999. The

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

645

Fig. 4. Accumulated return for each stock (in the order listed in Table 1) with di erent for Pc = 0:1 (a) and Pc = 0:05 (b).

(week to years)

return R is dened from the sell price (S ) and bought price (B), R = 100 (S B)=B : (6) The maximum and minimum returns are easy to locate (see Fig. 3). The accumulated return RA is dened as RA =
i

Ri ;

(7)

where Ri is the return (Eq. (6)) during ith trade. Fig. 4 shows the accumulated return (gain/loss) for each stock with threshold Pc = 0:05 and 0.10 for a range of periods (short to long time). The accumulated return

646

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

Fig. 5. Trade frequency of each stock with di erent for Pc = 0:10 (a) and Pc = 0:05 (b) during 19922002. The order of stock symbol on x-axis follows Table 1.

(Eq. (7)) is the sum of all returns if all possible trades during 19922002 were executed excluding the trading commission. It is easy to see that on the average the accumulated returns are higher with the lower periods ( ) of the momentum series. This is not unexpected, as the frequency of trade decreases generally with increasing the period of momentum series (see Fig. 5). Note that the numbers of trades are higher with lower trading threshold. The qualitative nature of maximum, minimum, and average returns among the di erent components of DJI remains very similar to that of accumulated return with varying uctuations.

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

647

With our momentum trading method, the frequency of trade is generally higher with smaller cycle of the time series (see Fig. 5(a) and (b)) with accordingly larger uctuations. It is interesting to note that for a long-time horizon with larger cycle ( 60 days), the frequency of trade is relatively constant among all the DJI components. Generally, the larger the threshold the fewer are the trades (cf. Fig. 5(a) and (b)). Selection of threshold depends on the frequency of trades and risk tolerance. E ects of trading threshold on each stock are also analyzed in detail. Fig. 6(ac) shows the variation of accumulated return with the threshold for each DJI component. It seems that for all stocks in general the accumulated return is higher when Pc lies between 0.04 and 0.06, and lowest when Pc is larger than 0.2. For all threshold below a certain value (Pc 6 0:1), the return is relatively constant with some uctuations in few stocks. At large threshold (Pc 0:2), the frequency of trade becomes very low (Fig. 7) with the lowest accumulated returns (Fig. 6). 4. Summary and conclusions A technical analysis of DJI stocks based on the historical data during 19922002 is presented with a momentum trading approach. Price change of each stock in a unit time ( ), normalized by the stock price denes the velocity u(t ) (Eq. (2)). The relative velocity v(t ) (Eq. (3)) is obtained by normalizing u(t ) by the maximum value uh (t ) among all 30 components during the span of unit cycle time ( ); a set of periods are used, = 5; 20; 30; 60; 120 and 240 days. This normalization uh (t ) is important factor to correlate relative speed of stocks prices within a sector, i.e., those in DJI. It varies with time and changes from one stock to another for t  . Normalized trade volume of each stock is the measure of its inertia (mass). The market momentum is the product of mass and relative velocity (Eq. (5)). The rate of momentum change is the driving force for trading and price change. The extrema in time series of momentum, i.e., the singularities in the driving force provide opportunities for trade which is controlled by the momentum trading threshold Pc . The threshold is an excellent tool for the limit orders (p(t ) Pc (sell), p 6 Pc (buy)). Returns are examined for a range of threshold Pc = 0:01 0.5 for the DJI data for a period of 10 years (19922002); the interesting range are found to be Pc =0:02 0.1. The probability of return is found to be high with such momentum trading based on these historical data. The choices of period ( ) for the momentum time series and trading threshold (Pc ) depend on the objectives (i.e., long to short term trading horizon). Although, all our analyses are based on the market data, the probability of return is not an exact science. This is an academic exercise to see that the basic law of classical dynamics in physics is applicable in stock analysis with some modications. Acknowledgements Support for visiting scholar program by Oversea Scholarship Program of UESTC (University of Electronic Science and Technology of China) is acknowledged by Hui

648

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

Fig. 6. (ac) Accumulated returns of DJI components versus Pc .

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

649

Fig. 7. (ac) Trade frequency of DJI components versus Pc .

Wang. Partial support from the Basic Research award by the University of Southern Mississippi is acknowledged by R.B. Pandey. We thank Johan Bjursell and Dietrich Stau er for useful comments and discussion.

650

H. Wang, R.B. Pandey / Physica A 331 (2004) 639 650

References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] Tradetrek University (2001), http://www.tradetrek.com. http://www.nance.yahoo.com. http://www.cnbc.com. R.W. Colby, The Encyclopedia of Technical Market Indicators, 2nd Edition, McGraw-Hill Trade, New York, October 22, 2002. N. Vandewalle, M. Ausloos, Ph. Boveroux, Physica A 269 (1999) 170. M. Ausloos, K. Ivanova, arXiv:cond-mat/0304700 v1 30, 2003. A.Y. Lo, The Econometrics of Financial Markets, Princeton University Press, Princeton, 1997. M. Ausloos, N. Vandewalle, P. Boveroux, A. Minguet, Physica A 274 (1999) 229. H.E. Stanley, L.A.N. Amaral, D. Canning, P. Gopikrishnan, Physica A 269 (1999) 156. J.P. Bouchaud, M. Potters, Theory of Financial Risks: From Statistical Physics to Risk Management, Alea Saclay, 1997. R.N. Mantegna, H.E. Stanley, An Introduction to Econophysics, Cambridge University Press, Cambridge, 1999. I. Chang, D. Stau er, R.B. Pandey, Int. J. Ther. Appl. Finance 5 (2002) 1. S. Focardi, C. Jonas, Modeling the Market: New Theories and Techniques, Frank J Fabozzi Association, 1997. P.W. Anderson, K.J. Arrow, The Economy as an Evolving Complex System, Addison-Wesley, Redwood City, 1988. S.M. de Oliveira, P.M.C. de Oliveira, D. Stau er, Evolution, Money, War, and Computers, Teubner, Leipzig, 1999. M. Ausloos, K. Ivanova, Eur. Phys. J. B 27 (2002) 177. M. Ausloos, K. Ivanova, arXiv:cond-mat/0212641 v1 31, 2002. K. Ivanova, L.T. Wille, Physica A 313 (2002) 625. J. Van Der Hart, E. Slagter, J. Empirical Finance 10 (2003) 105.