Chapter 7: NetWork Analysis and performance
To check the performance of ANN network There are several methods have been done to
examine data set, data preprocessing, Neural network paradigm/structure, Survivorship
Bias, walk forward and backward windows, (RMSD) root-mean-square deviation error or
(RMSE) root-mean-square error and Histogram for ERROR Frequencies.
The data set partitioned into three groups which are: Training set, validation, and Testing set.
I used 7-algorithms pre-processing to maintain the downloaded file before I feed it to
ANN to deleting unnecessary columns, Check that the length is not less the 100 day,
Add four columns at the end of data file, calculate technical Indicators to make sure that
the total inputs that were used are nine input , deleting any rows that contain constant
values, reduced and scale each trend output price between minimum and maximum
values by using log-sigmoid function and scale the trend output between to 0 (Down) and
+1 (Up), eliminated the first row from data file which contains header, and Finlay check
the file format if it was not in standard form.
The stocks utilized for this research were individually selected as widely held
International stocks over the testing period. Therefore, the research suffers from
survivorship bias which is the tendency to exclude failed companies from performance
It must be noted however that while some data may suffer from survivorship bias; this is
not considered a serious issue for the study. Survivorship bias is typically a problem in
studies examining mutual fund performance Where This study does not involve the
performance measurement of surviving mutual funds or stocks. This study uses ANNs to
predict the future prices of a group of different stocks.
The look back window indicates and had a place with what number of past data illustrates
and satisfied what the system is giving to make the prediction. I used a tapped Windows
as a part of the system; a look back window is made whereby the system is permitted to
utilize various past inputs to make the following expectation.
The ANN structure adopted consisted of 3-layer network which comprises the input
layer, a hidden layer, and the output layer. This system is a feed-forward with
propagation network. Heuristics and Method were used to estimate hidden layer size. The
output layer size is a relatively simple matter. The network was structured with a two
outputs neuron so that for each time step, the network provided an estimate of price at:
(2t +15) (the price and trend of the stock in fifteen weeks’ time).
(Minimum days or raw to accept data ;=100 row or days = 14.5 week =15).
Transfer functions is the Log-sigmoid transfer functions, this allowed for negative input
weights which was consistent with the trend input data (many of the inputs could take the
form of either positive (Up-trend) or negative values- (Down trend). The log-sigmoid
transfer function gives the output only (0 or 1).
The (RMSD) root-mean-square deviation error or (RMSE) root-mean-square error is a
frequently used to calculate the effective rate or successful measurement of a varying and
different set of values, and that was proved for equal and unequal daily periods from
2000 till 2016.
There are different models measures the performance. Such as: Fama French Factor,
Fama French 3 Factor Model plus Carhart’s “Momentum factor, QMJ 4 Factor Model.
Hedge Fund risk factor Model, Fama French’s New 5 Factor Model, Carhart four-factor
After applied Cahart four factor model on our ANN, this implies that the return of long
portfolio is less volatile than the daily or short income;
7.2. Data set:
The data set used for the analysis was obtained by using random stocks from the biggest
company which belonged to several stock exchanges such as:
? GE General Electric Company NYSE
? KMI Kinder Morgan, Inc NYSE
? KS Samsung Electro-Mechanics Co. Ltd. KSE
? MSFT Microsoft Corporation Nasdaq GS
7.1.1. Definition and meaning of some stock market exchanges:
? NASDAQ – NYSE:
The NASDAQ Stock Market, commonly known as the NASDAQ (currently stylized
as Nasdaq), is an American stock exchange. It is the second-largest exchange in the world
by market capitalization, behind only the New York Stock Exchange.
platform is owned by Nasdaq, Inc., which also owns the OMX stock market network and
several other US stock and options exchanges.
? NYSE Euronext:
NYSE Euronext, Inc. was a Euro-American multinational financial services corporation
that operated multiple securities exchanges, including the New York Stock
Exchange, Euronext and NYSE Arca (formerly known as ArcaEx). NYSE merged with
Archipelago Holdings on March 7, 2006, forming NYSE Group, Inc.2 On April 4,
2007, NYSE Group, Inc. merged with Euronext N.V. to form the first global equities
“Monthly Reports”. World-Exchanges.org. World Federation of Exchanges. Archived from the originalon August 17, 2014.
Retrieved June 3, 2015.
exchange, with its headquarters in Lower Manhattan.
The components were then part of
Intercontinental Exchange, although it has now spun off Euronext.
? Karachi Stock Exchange:
The Karachi Stock Exchange Limited (KSE), was a stock exchange located at the Stock
Exchange Building (SEB) on Stock Exchange Road, in the heart of Karachi’s Business
District, I. I. Chundrigar Road, Karachi, Sindh Province of Pakistan. It is now
incorporated in the Pakistan Stock Exchange along with the other two bourses of
Pakistan, the Lahore Stock Exchange and the Islamabad Stock Exchange
Pakistan’s largest and one of the oldest stock exchange in South Asia by market
capitalization, with many Pakistani consortium as well as overseas enterprises listings.
According to Bloomberg, the Pakistani benchmark stock market index is the third-best
performer in the world since 2009.
In June 2015, Khaleej Times reported that since
2009, the Pakistani equities delivered 26 percent a year for US dollar investors, making
Karachi the best-performing stock exchange in the world.
With effect from January 11, 2016 the Karachi Stock Exchange, Lahore Stock Exchange
and Islamabad Stock Exchange were integrated under the Stock Exchanges
(Corporatization, Demutualization and Integration) Act, 2012 to form the Pakistan Stock
Exchange Limited as the only stock exchange in Pakistan.
This group of stocks was used for all analysis undertaken. The data obtained using historical
data from International financial websites which have price information (open, high, low, and
close), volume, and several fundamental inputs which are discussed further below. Daily,
weekly and monthly data were obtained for price based inputs. The database is based upon
company financial statements included:
Pallavi Gogoi (February 15, 2011). “NYSE, Deutsche Boerse are already global companies”.Bloomberg Businessweek.
Retrieved February 5, 2012.
Chad Bray (May 27, 2014).”IntercontinentalExchange Set to Spin Off Euronext”. New York Times. Retrieved August 25, 2014.
News, Dawn. “Pakistan Stock Exchange formally launched”.
47 “Pakistan central bank cuts benchmark rate to 42-year low”. AFRWEEKEND. May 24, 2015. Archived from the original on May
24, 2015. Retrieved May 24,2015.
48 “KSE world’s best performing frontier stock market: report”. The Express Tribune. June 3, 2015. Archived from the original on Jun
8, 2015. RetrievedJun 8, 2015.
? 1942 daily data points (259 week=65 month=5.6 years) for each daily stock for the period
from 1/3/2010 (mm/dd/yyyy) to 7/15/2016 (equal periods equal one year for each and
unequal starting from 6 months,1year,1.5 year, two years…. to 6 years.
? 195 weekly data points (1463 day =49 month =4.1 years) for each daily stock for the
period from to 1/4/2010 (mm/dd/yyyy) to 1/5/2015.
? 438 monthly data points (13150 day=1753 week= 37.5 year) from 1/3/1980 to 1/4/2016 a
7.1.2. Survivorship Bias
The stocks utilized for this research were individually selected as widely held
International stocks over the testing period. Therefore, the research suffers from survivorship
bias which is the tendency to exclude failed companies from performance studies.
While survivorship bias makes research findings less robust, survivorship bias is the
widespread in the extant literature where ANNs are used for price prediction and portfolio
selection. For example, the papers by Ko and Lin (2008), Yoon and Swales (1991),
Kryzanowski et al. (1992) all appear to suffer from survivorship bias. There are however
some studies that are free from survivorship bias including Ellis and Wilson (2005) and
Vanstone et al. (2010).
It must be noted however that while some data may suffer from survivorship bias; this is not
considered a serious issue for the study. Survivorship bias is typically a problem in studies
examining mutual fund performance where the study includes only surviving funds in the
analysis. This is a problem because generally, funds that disappear do so due to poor
performance (Elton, Gruber ; Blake, 1996). It is well established that survivorship bias in
such performance studies weakens evidence of performance persistence (Brown, Goetzmann,
Ibbotson ; Ross, 1992; Carhart, Carpenter, Lynch ; Musto, 2002; Carpenter ; Lynch,
1999). This typical situation needs to be distinguished from this study. This study does not
involve the performance measurement of surviving mutual funds or stocks. This study uses
ANNs to predict the future prices of a group of different stocks.
This study is intended to determine if ANNs show sufficient ability to predict stock prices on
a small scale to justify further research on a larger scale. Given the paucity of published
research in the field, an exhaustive study that utilizes a wide investment universe that is free
of survivorship bias is considered premature until there is a better understanding of:
• Which network inputs achieve maximum network predictive capability?
• Reliable heuristics for network specification.
The task of creating a survivorship bias free data set for this type of ANN trading system will
be a difficult and complex exercise. Obtaining the inputs (particularly the fundamental
inputs) for a large universe of stocks including delisted stocks over a long-time horizon may
be difficult to locate and expensive to acquire.
Copyright 2019 - Education WordPress Theme.