Background

BACKGROUND

      The creation of a machine learning framework to perform stock market price prediction assumes that stock market price history tends to repeat itself. Basically, this means that a stock’s past price movements will be able to predict its future price movements. In economic terms, the assumption is that the market is “efficient.” Efficiency in the market means that the price of any individual security at any given point in time reflects its intrinsic value [1]. As such, according to the economist Fama, efficiency requires that stock prices be able to instantaneously adjust to the actual stock price [2]. Because of this, traditional economic theory would not classify the market as efficient.

In the view of traditional economic theory, the stock market is unstable and inefficient. This view is rooted in the idea that there is some vagueness and uncertainty about new, incoming stock information, meaning that there cannot be “instantaneous” adjustment. According to Fama, these adjustments have two major implications that disprove the idea of “market efficiency”. Firstly, actual prices will initially over-adjust in intrinsic value as often as they will under-adjust. Secondly, the lag in the complete adjustment of a stock’s value to its actual intrinsic value will be independent and random Thus, according to these two observations, successive prices changes in individual securities will be independent of past prices, which is the definition of what Fama called a “random walk.” Based on the ideas of the random walk and market efficiency, a series of stock price changes does not have any memory. Later on, Fama would expand these theories into his Efficient Market Hypothesis. Basically, a major implication of the EMH is that consistently comes at substantial financial risk [2].

LOOKING FOR MARKET PREDICTABILITY

However, research following Fama’s findings in the 1960s have shown that the market does have some level of predictability. One level of this predictability is the fluctuation of price between the bid and ask, which is known as the bid-ask spread. The bid price is defined as the price an individual is willing to pay for a stock, while the ask is the price at which an individual is willing to sell a stock. The true value of the stock lies somewhere between the bid and ask [3]. In an efficient market, the bid and ask fluctuate randomly. However, according to Timmerman and Grange, the existence of a single successful trading model would be sufficient to demonstrate a violation of the EMH [4].

There are several studies indicating success at using pricing data in order to predict future price movements. Phua et al utilized neural networks to predict the movement of five major stock indices: DAX, DJIA, FSTE-100,HIS and NASDAQ. Their rate of accurate prediction using only stock prices as inputs exceeded about 60% [5]. In his study, Kim applied a support vector machine (SVM) to predicting daily Korean stock market prices, yielding an accuracy of prediction of about 56% [6]. Huang et al applied backpropagation neural techniques to the Japanese NIKKEI index, achieving predictive accuracy greater than 70% [7]. These successful trading models would support Timmerman and Granger’s theory that the market does have some time periods of predictability.

Timmerman and Granger extrapolated on time periods where market predictability could be found. According their work, if there are instances of high probability patterns during a time-span, they will be more likely to be spotted over time and manipulated by traders in their trading strategy. Yet, this widespread adoption of similar approaches to trading strategy would likely increase or decrease stock prices enough that the pattern would be elimination, making the price randomized. Such a line of logic does leave room for potential “hot spots,” where prediction methods are possible to implement. Timmerman and Grange suggest possible procedures to find these “hot spots” by using wide searches across both models that adapt quickly and a large number of assets. They conclude that traders who first implement new financial prediction methods (before the pattern is eliminated by widespread adoption) are the most likely to be successful, leading to the necessity for what Timmerman and Grange call “the race for innovation.”

 

 

Skip to toolbar