International Research Journal of Finance and Economics ISSN 1450-2887 Issue 43 (2010) © EuroJournals Publishing, Inc. 2010 http://www.eurojournals.com/finance.htm Inefficiency in Gold market Mohsen Mehrara Kargar-e-Shomali Avenue, Faculty of Economics, University of Tehran, Tehran, Iran Tel: +98-21-88029007 E-mail: mmehrara@ut.ac.ir Ali Moeini Address: Enghelab Street, Department of Algorithms and Computation, Faculty of Engineering, University of Tehran, Tehran, Iran Tel: +98-21-66406186 E-mail: moeini@ut.ac.ir) Mehdi Ahrari Address: Kargar-e-Shomali Avenue, Faculty of Economics, University of Tehran, Tehran, Iran Tel: +98-21-88029007 E-mail: ahrari@ut.ac.ir Vida Varahrami Address: Kargar-e-Shomali Avenue, Faculty of Economics, University of Tehran, Tehran, Iran Tel: +98-21-88029007 E-mail: vida7892000@yahoo.com, vvarahrami@ut.ac.ir Abstract The efficient market hypothesis gives rise to forecasting tests that mirror those adopted testing the optimality of forecast in given information set. It seems obvious that profitable trading opportunities may exist, when the gold market is not efficient. In this paper, among the models such as GMDH and MLFF neural network, Technical analysis rule, Buy and hold rule, Treasury bill tested and Random walk rule, GMDH and MLFF neural network models with moving crossover inputs are determined to produce better results in the gold market. Results also indicate that high returns are possible by using GMDH neural network and at the end, we show that the gold market is not efficient in Fama sense. Keywords:
Multi Layered Feed Forward (MLFF) and Group Method of Data Handling
(GMDH) neural network, Gold prices, Technical Analysis. 1. Introduction Allan Timmermann and Clive W. J. Granger[41] note in their paper " Transaction costs and trading restrictions change tests of market efficiency in some important ways. Most obviously, if transaction costs are very high, predictability is no longer ruled out by arbitrage, since it would be too expensive to take advantage of even a large, predictable component in returns." They note in other section of this paper" The earlier definition of market efficiency is purely based on asset returns independently of how International Research Journal of Finance and Economics - Issue 43 (2010)
59
these are related to the underlying ‘intrinsic’ asset value. In fact, prices and values need not be closely related. This definition of market efficiency focuses on the size of deviations of asset prices from true ‘value’. Investors’ information can be so ‘noisy’ at times that price are far removed from fundamentals. However, the market is still efficient in this definition provided that such deviations do not last ‘too long’ or become ‘too big’ before they are corrected." Jensen [21] in his paper notes "A market is efficient with respect to information set
t if it is
impossible to make economic profits by trading on the basis of information set
t" and Malkiel [26]
notes " A capital market is said to be efficient if it fully and correctly reflects all relevant information in determining security prices. Formally, the market is said to be efficient with respect to some information set,
t, if security prices would be unaffected by revealing that information to all
participants. Moreover, efficiency with respect to an information set,
t, implies that it is impossible to
make economic profits by trading on the basis of
t".
Gold markets are a zero sum game, where for every long position there must be a short position. Every price movement in market that creates a profit for one participant will result in an equal loss for another participant. With heavy participation of informed traders on both sides, trading should perform particularly well as a price discovery mechanism based on the known market fundamentals. Thus neoclassical economic theory suggests that markets should be highly efficient, with no room for excess returns. What then explains traders' enthusiasm for short-term bets on prices? [36] Long before the contributions of behavioral finance, technical analysts have claimed the ability to generate excess returns solely on the basis of interpretations of historical price patterns. Technical analysis has evolved from looking for patterns on price charts to using computer programs based on a variety of weighted and unweighted price averages, to the current state of the art which employs artificial intelligence [AI] techniques. A branch of AI technology that has been experiencing strong expansion in its application to commodity trading is the field of GMDH and MLFF neural networks. If historical price patterns can indeed be used to predict future prices, the market would not fit the strict definition of efficient markets laid out by Fama [10]. If inefficiencies exist, there could be profitable trading opportunities. Problems of complex objects modeling such as analysis and prediction of stock market, gold price and other, cannot be solved by deductive logical-mathematical methods with needed accuracy. The task of knowledge extraction from data is to select mathematical description from data. But the required knowledge for designing of mathematical models or architecture of neural networks is not at the command of the users [27]. In mathematical statistics it is need to have a priori information about the structure of the mathematical model. In neural networks the user estimates this structure by choosing the number of layers, and the number and transfer functions of nodes of a neural network. This requires not only knowledge about the theory of neural networks, but also knowledge of the object's nature and time. Besides this the knowledge from systems theory about the systems modeled is not applicable without transformation into an object in the neural network world. But the rules of translation are usually unknown [18, 19]. GMDH type neural networks can overcome these problems. It can pick out knowledge about object directly from data sampling. The GMDH is the inductive sorting-out method, which has advantages in the cases of rather complex objects, having no definite theory, particularly for the objects with fuzzy characteristics. GMDH algorithms found the only optimal model using full sorting-out of model-candidates and operation of evaluation of them, by external criteria of accuracy or difference types [17, 24]. The purpose of this paper is to examine the possibility that the gold market is not an efficient market in that subtle price patterns that can be exploited for profitable trading. To explore this question we estimate GMDH and MLFF neural network model for the price of nearby gold with technical analysis rules as inputs. We find that, without transaction costs and disregarding slippage, the GMDH model can produce profitable trading signals for several years, thus casting doubt on the efficiency of the gold market. 60
International Research Journal of Finance and Economics - Issue 43 (2010)
In Section 2 below, we reference selected literature as background, including studies which have used neural network models in technical analysis. Section 3 provides the non-parametric modeling approach adopted here as per MLFF neural network with back-propagation algorithm and GMDH neural network with genetic algorithms which are briefly discussed and describes a network with technical analysis rules as inputs. Benchmark models are given in Section 4. Empirical results are presented in Section 5, and concluding remarks in Section 6. 2. Literature Survey Gencay (13) uses foreign exchange markets to pioneer the use of technical analysis rules as inputs for neural networks, which are flexible, nonlinear models with powerful pattern recognition properties. In a series of articles, Gencay (14) and Gencay (15) and Gencay et al. (16) show that simple technical rules result in significant forecast improvements for current returns over a random walk model for both foreign exchange rates and stock indices. Doran [9] employing a survey distributed to over 4,000 professors, they obtain four main results. First, most professors believe the market is weak to semi-strong efficient. Second, twice as many professors passively invest than actively invest. Third, their respondents’ perceptions regarding market efficiency are almost entirely unrelated to their trading behavior. Fourth, the investment objectives of professors are, instead, largely driven by the same behavioral factor as for amateur investors–one's confidence in his own abilities to beat the market, independent of his opinion of market efficiency. Sohel Azad [37] empirically tests the random walk and efficiency hypothesis for 12 Asia- Pacific foreign exchange markets. The hypothesis is tested using individual as well as panel unit root tests and two variance-ratio tests. The study covers the high (daily) and medium (weekly) frequency post-Asian crisis spot exchange rate data from January 1998 to July 2007. The inferential outcomes do not differ substantially between the unit root tests and the variance-ratio tests when using daily data but differ significantly when using weekly data. The use of gold price as an economic indicator drew the attention of many economists. Mirmirani [27] asserted that more than 70% of the change in inflation rate can be explained by the price movements of gold. In the field of engineering decision systems, techniques based on inductive human behavior have been developed. Among such systems are neural networks, the genetic algorithm and their integration. These engineering-based systems have gradually found their way into business and economics [8]. A thorough literature review of neural network applications in finance and business are provided by Wong and Yakup [44]. In time series analysis, a review of the methodological linkage between statistical techniques and neural networks is given by Cheng and Titterington [7]. Recently, some studies have been done with neural networks. One of them has been done with S. Mirmirani and H.C. LI [27], concentrates on a non-parametric study of gold prediction with neural network and genetic algorithm. Sarafraz and.Afsar [35] has done another study with neuro-fuzzy networks. They have studied gold price in Iran. 3. Modeling using Neural Network Artificial Neural Networks (ANN) is biologically inspired network based on the organization of neurons and decision making process in the human brain [26]. In other words, it is the mathematical analogue of the human nervous system. This can be used for prediction, pattern recognition and pattern classification purposes. It has been proved by several authors that ANN can be of great used when the associated system is so complex that the underline processes or relationship are not completely understandable or display chaotic properties [31]. Development of ANN model for any system involves three important issues: (i) topology of the network, (ii) a proper training algorithm and (iii) transfer function. Basically an ANN involves an input layer and an output layer connected through one International Research Journal of Finance and Economics - Issue 43 (2010)
61
or more hidden layers. The network learns by adjusting the inter connections between the layers. When the learning or training procedure is completed, a suitable output is produced at the output layer. The learning procedure may be supervised or unsupervised. In prediction problem supervised learning is adopted where a desired output is assigned to network before hand [20]. 3.1. MLFF Neural Network MLFF neural network is one the famous and it is used at more than 50 percent of researches that are doing in financial and economy field recently [1]. This class of networks consists of multiple layers of computational units, usually interconnected in a feed-forward way. Each neuron in one layer has directed connections to the neurons of the subsequent layer. In many applications the units of these networks apply a sigmoid function as a transfer function ( x
e
f x
−
+
=
1
1
( )
). It has a continuous derivative,
which allows it to be used in back-propagation. This function is also preferred because its derivative is easily calculated:
y '= y (1− y )
Multi-layer networks use a variety of learning techniques; the most popular is back-propagation algorithm (BPA). The BPA is a supervised learning algorithm that aims at reducing overall system error to a minimum [22]. This algorithm has made multilayer neural networks suitable for various prediction problems. In this learning procedure, an initial weight vectors
0 w is updated according to
[21]: i i i i i i i
w
(k +1) = w (k ) + μ (T − O ) f ' (w x )x (1)
Where,
i
w
The weight matrix associated with ith neuron;
i
x
Input of the ith neuron;
i
O
Actual output of the ith neuron;
i
T
Target output of the ith neuron, and μ is the learning rate
parameter. Here the output values (
i O ) are compared with the correct answer to compute the value of some
predefined error-function. The neural network is learned with the weight update equation (1) to minimize the mean squared error given by [23]: 2 2
[ ( )]
2 1 ( ) 2 1 i i i i i
E
= T −O = T − f w x
By various techniques the error is then fed back through the network. Using this information, the algorithm adjusts the weights of each connection in order to reduce the value of the error function by some small amount. After repeating this process for a sufficiently large number of training cycles the network will usually converge to some state where the error of the calculations is small. In this case one says that the network has learned a certain target function. To adjust weights properly one applies a general method for non-linear optimization that is called gradient descent. For this, the derivative of the error function with respect to the network weights is calculated and the weights are then changed such that the error decreases [24]. The gradient descent back-propagation learning algorithm is based on minimizing the mean square error. An alternate approach to gradient descent is the exponentiated gradient descent algorithm which minimizes the relative entropy [38]. 3.2. GMDH Neural Network By means of GMDH algorithm a model can be represented as sets of neurons in which different pairs of them in each layer are connected through a quadratic polynomial and thus produce new neurons in the next layer. Such representation can be used in modeling to map inputs to outputs [3, 12, 17, 18, 25]. The formal definition of the identification problem is to find a function
f ˆ so that can be approximately
used instead of actual one,
f , in order to predict output y ˆ for a given input vector ,..., )
3 , 2 , 1 ( n X
= x x x x
as close as possible to its actual output
y [19,28]. Therefore, given M observation of multi-inputsingle-
output data pairs so that 62
International Research Journal of Finance and Economics - Issue 43 (2010)
,..., ) 3 , 2 , 1 ( in x i x i x i f x i y
= (i=1,2,…,M)
it is now possible to train a GMDH-type neural network to predict the output values i y
ˆ for any given
input vector
( , , ,..., )
i
1 i 2 i 3 in
X
= x x x x , that is
,..., ) 3 , 2 , 1 ˆ ˆ ( in x i x i x i f x i y
= (i=1,2,…,M).
The problem is now to determine a GMDH-type neural network so that the square of difference between the actual output and the predicted one is minimized, that is min 2 ,..., ) ] 3 , 2 , 1 ˆ ( 1
[ − ®
= i y in x i x i x i f x M i . General connection between inputs and output variables can be expressed by a complicated discrete form of the Volterra functional series in the form of ... 1 1 1 1 1 1 0 = =
+
= = =
+ +
= = + n i n j n i n j x j xk n k xi aij xi x j aijk xi n i y a ai
(2)
Which is known as the Kolmogorov-Gabor polynomial [3, 18, 19]. This full form of mathematical description can be represented by a system of partial quadratic polynomials consisting of only two variables (neurons) in the form of 2 5 2 y
ˆ = G (xi , x j ) = a 0 + a 1xi + a 2 x j + a 3 xi x j + a 4 xi + a x j (3)
In this way, such partial quadratic description is recursively used in a network of connected neurons to build the general mathematical relation of inputs and output variables given in equation (2). The coefficient
i a in equation (3) are calculated using regression techniques so that the difference
between actual output, y, and the calculated one,
y ˆ , for each pair of i x , j x as input variables is
minimized. Indeed, it can be seen that a tree of polynomials is constructed using the quadratic form given in equation (3) whose coefficients are obtained in a least-squares sense. In this way, the coefficients of each quadratic function i G
are obtained to optimally fit the output in the whole set of
input-output data pair, that is [18, 28] 1 min ( )2 ® = − = M M i i G i y E
(4)
In the basic form of the GMDH algorithm, all the possibilities of two independent variables out of total n input variables are taken in order to construct the regression polynomial in the form of equation (3) that best fits the dependent observations ( i y
, i=1, 2, …, M) in a least-squares sense.
Consequently, 2 ( 1) 2 − =
n n n
neurons will be built up in the first hidden layer of the feed forward network from the observations {
( , , );
iq x ip yi x
(i=1, 2,… M)} for different p ,q Î{1,2,...,n } . In other
words, it is now possible to construct M data triples { (
yi , xip , xiq ); (i=1, 2,…, M)} from observation
using such
p ,q Î{1,2,...,n } in the form
Mp Mq M p q p q x x y x x y x x y 2 2 2 1 1 1 International Research Journal of Finance and Economics - Issue 43 (2010)
63
Using the quadratic sub-expression in the form of equation (3) for each row of M data triples, the following matrix equation can be readily obtained as A
a = Y
where
a is the vector of unknown coefficients of the quadratic polynomial in equation (3)
{ , , , , , }
0 1 2 3 4 5 a = a a a a a a (5)
and
T
M
Y
{y , y , y ,..., y } 1 2 3 = is the vector of output’s value from observation. It can be readily seen that
= 2 2 2 2 2 2 2 2 2 2 2 1 2 1 1 1 1 1 1 1 1 Mp Mq Mp Mq Mp Mq p q p q p q p q p q p q x x x x x x x x x x x x x x x x x x A The least-squares technique from multiple-regression analysis leads to the solution of the normal equations in the form of A A A Y
T
1 T ( )− a = (6)
which determines the vector of the best coefficients of the quadratic equation (3) for the whole set of M data triples. It should be noted that this procedure is repeated for each neuron of the next hidden layer according to the connectivity topology of the network. However, such a solution directly from normal equations is rather susceptible to round off errors and, more importantly, to the singularity of these equations [2, 5, 29, 30]. Our approach is to construct GMDH and MLFF neural networks model using a technical analysis rule as an input. We assume that technical analysis rules may have merit because there may be exploitable patterns which can result in profitability. The patterns may not be identifiable by traditional means but may be uncovered by the pattern recognition capabilities of GMDH neural networks. Empirical evidence of profitability will be confirmation of this approach and evidence of inefficiencies in the gold market. 2 lags of the 5[MA
5,MA5(-1),MA5(-2)], 50[MA50,MA50(-1),MA50(-2)], day moving average
crossover
1, as input variables to the neural networks. Moving average crossovers have long been used
as buy– sell signals for trend-following trading systems. The assumption that a short-term moving average is higher than a long-term moving average indicates that prices are trending higher, thus signaling an uptrend. The opposite case signals a downtrend. First we use the neural network with moving average crossover inputs to forecast what will happen to gold prices. A long position is taken when prices are predicted to be higher, and a short position when prices are predicted to be lower. Then profitability is calculated from actual returns based on these positions. For tractability, we utilize a neural network with two hidden nodes and a direct connection between the lagged moving average crossovers and return. The price data are NYMEX database. Observations are the daily nearby gold prices. Predicted prices from the model are compared with the actual prices for 01/01/2006 until 31/6/2009 period separately. Prices are converted to returns using the daily difference in the logs of the prices. 4. Other Trading Strategies In this paper, in addition to the GMDH and MLFF neural network, three benchmark models are used for comparison purposes. The first benchmark model, referred to as the buy-and-hold [BH] model, simply assumes that the nearby contract is purchased and rolled over as a new contract month becomes the nearby contract. The second benchmark model [TA for “twenty-day average”] is a simple moving 1
Such models are all based on rules using moving averages of recent prices. A typical moving average is simply the sum
of the closing prices for the last
n number of days divided by n, where n may be from 1 to 200 days the rules for using
these tools are very similar and usually involve making a decision when a short-term average crosses over a long-term average. For example, the rule may be to buy when the 5-day moving average exceeds the 50-day moving average and to sell when the 5-day average is below the 50-day average.(Gencay et al. 1996) 64
International Research Journal of Finance and Economics - Issue 43 (2010)
average crossover model where a long position is established when the price exceeds the 20-day moving average by one standard deviation, and a short position is taken when the price is at least one standard deviation below the 20-day moving average. This second model has been popular with technical analysts, who believe it has the ability to capture and profit from trending markets. The third model [RW for “random walk”] incorporates a naïve trading rule based on the current day's movement. If the price is above the previous day's closing price, the trader goes long. If the price is below the previous day's price, the trader goes short, for more details referred to Shambora and Rositer [36]. In Table 1, we summarize the neural network given in Section 3 as well as the benchmark trading strategies discussed in this section. Table 1:
Summary of trading strategies
Model Description Strategy GMDH & MLFF neural network Inputs to the network are the differences between the 5- and the 50-day moving average of price (1–5 lags) Take a long position if forecast from the network is that price will increase; take a short position if forecast is that price will decrease. TA; Technical analysis rule Conventional moving average crossover rule Take a long position if price exceeds the 20-day moving average by one standard deviation; take a short position if price is at least one standard deviation below the 20-day moving average. BH; Buy and hold rule Assumes price will always increase Go long and continuously roll position over. T-bill 90-day; Treasury bill Risk-free rate Measure equal to change in log price of 90-day Treasury bill RW; Random walk rule Naïve trading rule Take a long position if price is above previous day's price; take a short position if price is below previous day's price. 5. Empirical Results We use trading signals from the four models to generate daily profit and loss expressed as percent returns and summarize the returns for each strategy in Table 2. Table 2:
Comparison of profitability
Model 2006 2007 2008 2009 Bear Bull Cumulative GMDH 0.675 0.689 0.623 0.594 0.376 0.612 2.581 MLFF 0.631 0.642 0.568 0.526 0.298 0.537 2.367 TA 0.456 0.523 0.496 0.437 0.228 0.453 1.912 BH 0.231 0.213 0.275 0.356 -0.237 0.654 1.075 RW -0.102 0.032 0.125 -0.117 -0.215 -0.0231 -0.508 T-bill 0.031 0.032 0.046 0.066 0.024 0.062 0.175 As shown in table 2 cumulative returns for GMDH and MLFF are 258% and 232%, but for TA, BH, RW and T-bill are 191%, 108%, -51% and 18%, then GMDH and MLFF neural network generate overall returns much better than other models and GMDH model is the most profitable. In this paper, we define period from Jun 10, 2007 to February 14, 2008 as a bear market because prices were generally falling in this period and the period from February 14, 2008 to December 31, 2008 is defined as a ball market which prices generally rising thru this period. Results show that return of BH model in the ball market is 65%, but return of this model in bear market is - 24%. Returns of GMDH and MLFF in bull market are lower than BH but in bear market returns of these models are higher than other models. International Research Journal of Finance and Economics - Issue 43 (2010)
65
In table 3 we compare average daily returns of 6 models. The average daily returns of GMDH and MLFF are 26% and 23% that are higher than other models. We use volatility for comparing returns. Volatility is measured with standard deviation of returns. In table 3 standard deviation of GMDH, MLFF and TA are quite high relative to other models; coefficient of variation is used for measuring comparative risk / reward. When there is small risk, CV is low. In table 3, the lowest CV and the highest risk is for GMDH model. Sharp ratio is another method for comparing returns relative to risk. The sharp ratio for GMDH is higher than other models; therefore neural network produces much better return for risk than other models. The t-static is identified for null hypothesis that average daily returns are significantly zero for overall period. Neural networks and T-bills' returns are significantly different from zero at 5% level. Table 3:
Statistical Comparisons 2006-2009
GMDH MLFF TA BH RW T-bill Average daily returns (%)
0.26 0.23 0.14 0.12 -0.046 0.01
Standard deviation
0.0278 0.0285 0.0229 0.0189 0.0092 0.0017
Coefficient of variation
10.69 12.39 16.36 15.75 -20 17
Sharpe ratio
0.0823 0.0778 0.0532 0.0485 -0.0345 NA
t-Stat
2.56* 2.34* 1.72 1.65 1.27 2.45*
Correct Trades (%)
66.4 63.7 62.3 53.2 48.1 NA
* Significant at 5% level. Correct trader is presented in table 3. GMDH and MLFF are correct 66.4% and 63.7% of the time that they are better than other strategies. In table 4, we compare correct trades and overall returns for three different opportunity costs. When opportunity cost increases, correct trades increases and overall return decreases. Overall results are shown that GMDH neural network dominates the other trading strategies Table 4:
Effect of opportunity cost filter trading rule on GMDH
Opportunity cost % Correct Trades Overall return 0.005% 77.23 2.28 0.01% 79.68 1.75 0.015% 83.35 1.21 7. Conclusion Efficient market theory claims that the efficient use of all available information are reflected by current prices. Strictly speaking, if markets are efficient, exploitable historical price patterns should not exist. Technical analysis, on the other hand, is the ability to identify patterns and profit from their repetition. Neural networks are especially well-suited to identifying patterns and making predictions based on those patterns. In this paper, we trained GMDH and MLFF neural network model for nearby gold prices with technical analysis crossover rules as inputs. By contrasting this model with a buy-and-hold (BH) strategy, a traditional technical trading (TA) strategy, a naïve “random walk” (RW) strategy and the return on Treasury bills (T-bill), we show that high returns are possible by using GMDH neural network. Overall returns, year-to-year returns, returns over a market cycle, and Sharpe ratios all favor the GMDH neural network model by a large factor. Additionally, daily returns using the neural networks and TA models are significantly different than zero, while the returns from the other models are not. There for the gold market is not efficient in the Fama sense. 66
International Research Journal of Finance and Economics - Issue 43 (2010)
References [1] Akgiray, Vedat, Booth, G. Geoffrey, Hatem, John J., & Mustafa, Chowdhury; Conditional dependence in precious metal prices. Financial Review, 26, 367-386, 1991. [2] Amanifard N., Nariman-Zadeh N., Borji M., Khalkhali A. and Habibdoust A., Modelling and Pareto optimization of heat transfer and flow coefficients in microchannels using GMDH type neural networks and genetic algorithms; Volume 49, Issue 2, February 2008, Pages 311-325. [3] Anastasakis, N. Mort. "The Development of Self-Organization Techniques in Modeling: A Review of The Group Method of Data Handling (GMDH)", Department of Automatic Control & Systems Engineering The University of Sheffield, Mappin St, Sheffield, No. 813, 2001. [4] Ashenwald, Janette, Fink, Stefan, Tappeiner, Gottfried, 2002. Brave new modeling: cellular automata and artificial neural networks for mastering complexity in economics. Complexity 7 (1), 39–46. [5] Atashkari K., Nariman-Zadeh N., Gölcü M., Khalkhali A. and Jamali A.; Modelling and multiobjective optimization of a variable valve-timing spark-ignition engine using polynomial neural networks and evolutionary algorithms; Energy Conversion and Management, Volume 48, Issue 3, March 2007, Pages 1029-1041. [6] Brauner, E.O.; Dayhoff, J.E.; Xiaoyun Sun; Hormby, S.; Neural network training techniques for a gold trading model;Computational Intelligence for Financial Engineering (CIFEr), 1997., Proceedings of the IEEE/IAFE 1997 Volume , Issue , 23-25 Mar 1997 Page(s):57 – 63 [7] Cheng, Bing and Titterington, D.M; 1994, Neural networks: A review from a statistical perspective; Statistical Science, 9 (1), 2-3. [8] Dawid, Herbert; 1996, Adaptive Learning by Genetic Algorithms, Springer, Berlin [9] DoranJ, Peterson D, Wright C, Confidence opinions of market efficiency and investment behavior of finance professors, Journal of financial markets, 2009 [10] Fama, Eugene, 1970. Efficient capital markets: a review of theory and empirical work, Journal of Finance 25, 383–417. [11] Fang Chen, Jiuping Xu. "Factor analysis for well-off construction based on GMDH", World Journal of Modeling and Simulation, Vol. 2, No. 4, pp. 213-221, 2006. [12] Farlow, S.J., ed.; Self-organizing Method in Modelling: GMDH type algorithm; Marcel Dekker Inc., 1984 [13] Gencay, Ramazan, 1996. Non-linear prediction of security returns with moving average rules. Journal of Forecasting 15 (3), 165–174. [14] Gencay, Ramazan, 1998a. The predictability of security returns with simple technical trading rules. Journal of Empirical Finance 5 (4), 347–359. [15] Gencay, Ramazan, 1999.Linear, non-linear and essential foreign exchange rate prediction with simple technical trading rules. Journal of International Economics 47 (1), 91–107. [16] Gencay, Ramazan, Stengos, Thanasis, 1998. Moving average rules, volume and the predictability of security returns with feedforward networks. Journal of Forecasting 17 (5–6), 401–414. [17] Ivakhnenko, A.G. The group method of data handling; a rival of the method of stochastic approximation. Soviet Automatic Control, 13(3), 43-55, 1968. [18] Ivakhnenko, G. A. Ivakhnenko. "Problems of Further Development of the Group Method of Data Handling Algorithms." Part I, Pattern Recognition and Image Analysis, Vol. 10, No. 2, pp. 187–194, 2000. [19] Ivakhnenko. Problems of further development of the group method of data handling algorithms. Pattern Recognition and Image Analysis, 187-194, 2000. [20] Jang, J. R. and Sun, C.; Nero Fuzzy Modelling and Control; Proc. of the IEEE, P.P: 378-405, 1995 International Research Journal of Finance and Economics - Issue 43 (2010)
67
[21] Jensen M, Some anomalous evidence regarding market efficiency. Journal of Financial Economics 6, pp. 95–101, 1978. [22] Kamarthi, S.V. and Pittner, S; accelerating neural network training using weight extrapolation; Neural Network, 12, 1285-1299, 1999. [23] Kartalopoulos, S.V.; Understanding Neural Networks and Fuzzy Logic- Basic Concepts and Applications; Prentice Hall, New-Delhi, 2000. [24] Lippmann, R.P.; An introduction to computing with neural nets; IEEE Mag. 3 (4), P.P: 4-22, 1987 [25] Madala, H.R. and Ivakhnenko, A.G.; 1994, Inductive Learning Algorithms for Complex Systems Modeling; CRC Press Inc., Boca Raton, p.384. [26] Malikle B, Efficient market hypothesis. In: New Palgrave Dictionary of Money and Finance, Macmillan, London 1992. [27] Mmirmirani S. and H.C. Li; 2004, Gold Price, Neural Networks and Genetic Algorithm; Computational Economics 23: 193–200. [28] Muller and G.A. Ivakhnenko, “Recent developments of self-organizing modeling in prediction and analysis of stock market”, available in URL address: http://www.inf.kiev.ua/GMDHhome/articles/, 1996. [29] Nariman-Zadeh; A Darvizeh; G R Ahmad-Zadeh. “Hybrid genetic design of GMDH-type neural networks using singular value decomposition for modeling and prediction of the explosive cutting process”, Journal of Engineering manufacture Proceedings of the I MECH E Part B, Volume: 217 Page: 779 -- 790, 2003. [30] Nariman-zadeh, N.; Darvizeh, A.; Darvizeh, M.; Gharababaei, H “Modeling of explosive cutting process of plates using GMDH-type neural network and singular value decomposition”. Journal of Materials Processing Technology, 2002, vol. 128, no. 1-3, pp. 80-87, Elsevier Science. [31] Priestley M. B.; Non-linear and non-stationary time series analysis; Academic Press, 1988. [32] Ripley, Brian D.; Statistical aspects of neural networks. In J. Barndorff-Nielsen, L. Jensen and W.S. Kendall (eds.), Networks and Chaos – Statistical and Probabilistic Aspects, 40-123., 1993 Chapman & Hall, London. [33] Sanchez E., Shiba T., Zadeh L.A., “Genetic Algorithm and Fuzzy Logic System”, Word Sciencem, 1997. [34] Sanders, Dwight R., Boris, Keith, Manfredo Mark, 2004. Hedgers, funds, and small speculators in the energy futures markets: an analysis of the CFTC's Commitments of Traders reports, Energy Economics 26 (3), pp. 425–445. [35] Sarfaraz, Leyla and Afsar, Amir; 2007, A study on the factors affecting gold price and a neurofuzzy model of forecast; (2005): Published in: Tarbiat Modaress Economic Reasearch Journal 16. [36] Shambora William E. Rossiter R, "Are there exploitable inefficiencies in the futures market for oil?" Journal of Energy Economics, Volume: 29 Pages: 18-27, 2007. [37] Sohel Azad A.S.M, Random walk and efficiency tests in the Asia-Pacific foreign exchange markets: Evidence from the post-Asian currency crisis data, Reasearch in international business and finance, vol. 23(3), 322-338, 2009. [38] Srinivasan, N., Ravichandran, V., Chan, K.L., Vidhya, J.R., Ramakirishnan, S., Krishnan, S.M.; Exponentiated back propagation algorithm for multilayer feedforward neural networks; Neural Information Processing, 2002. ICONIP apos; 02. Proceedings of the 9th International Conference on Volume 1, Issue , 18-22 Nov. 2002 Page(s): 327 - 331 vol.1 [39] Swanson, N., White, H., 1995.A model-selection approach to assessing the information in term structure using linear models and artificial neural networks. Journal of Business & Economic Statistics 13, 265–275. 68
International Research Journal of Finance and Economics - Issue 43 (2010)
[40] Timmerman A, Granger C, Efficient market hypothesis and forecasting. International Journal of Forecasting Elsevier, vol. 20(1), 15-27, 2004. [41] Vasechkina, E.F. and Yarin, V.D., “Evolving polynomial neural network by means of genetic algorithms: some application examples”. Complexity International, Vol. 9, 2001. [42] Water P.R., S. Wibier, E.J.H. Kerckhoffs and H. Koppelaar, “GMDH-based stock price prediction”, Neural Network World, vol.7, no.4-5, pp.552-563, 1997. [43] Wong, Bo K. and Yakup, Selvi; 1998, Neural network applications in finance: A review and analysis of literature (1990–1996); Information and Management, 129–139.