SlideShare ist ein Scribd-Unternehmen logo
1 von 38
Downloaden Sie, um offline zu lesen
Wavelet Analysis
By Sam Luxenberg
23rd March 2017, Tel Aviv
INTRODUCTION TO WAVELETS
Applications:
▪ Data Compression
▪ Signal Analysis
▪ Smoothing/De-noising Data
▪ Modeling Abrupt Changes
▪ Pattern Recognition
▪ Solutions to Partial Differential Equations
WAVELETS AND FINANCIAL MARKETS
▪Markets are complex chaotic systems that have self-similarity properties.
▪Financial signals can be thought of as fractal signals having self-similarity properties
which allows for analysis with wavelets.
SIGNAL PROCESSING AND ANALYSIS
▪ The conventional tool is Fourier Analysis which can represent a signal (e.g. audio
signal, images, seismic signal, financial signal) as a sum of sinusoids (think cosine and
sine curves).
▪ We may be interested in certain cyclic or frequency components present in a signal.
▪ The Fourier Transform allows us to study the signal not just as it relates to time, but
also these frequency components.
REPRESENTATION OF SIGNALS
▪ In Fourier Analysis, a signal g(t) can be represented as:
Where are called the Fourier coefficients (aka Fourier
Transform of g(t))
▪ In Wavelet Analysis, a signal g(t) can be represented as:
Where are called the wavelet coefficients of g(t) with
respect to the basis
SIGNAL PROCESSING AND ANALYSIS
▪ Consider the Discrete Fourier Transform (DFT)
Let x0, x1, …, xN-1 be a signal sampled at N points in time.
Then the DFT is: where
▪Note that this is a linear transformation of the signal and can therefore be
represented as a matrix.
▪ This sum can be thought of as the correlation between the signal xn and the different
frequency components.
SIGNAL PROCESSING AND ANALYSIS
▪ Consider the Discrete Wavelet Transform (DWT)
Let x0, x1, …, x2n-1 be a signal sampled at 2n points in time.
Then via the DWT, x has its wavelet decomposition of the form:
where is a scaling function.
▪ Note that the DWT is also a linear (and orthogonal) transformation and can
therefore be represented as a matrix.
GENERAL STRUCTURE OF WAVELET ANALYSIS
2 Steps
▪ Decomposition with the DWT
▪ Reconstruction/Synthesis with the Inverse DWT (IDWT)
ONE LEVEL OF WAVELET DECOMPOSITION
▪ The DWT of a signal x is calculated by passing it through a series of filters.
▪ Samples are simultaneously passed through a lower-pass and high-pass filter
resulting in a convolution of the two.
▪ (approximation or scaling coefficients)
▪ (detail coefficients)
ADVANTAGES OVER FOURIER TRANSFORM
▪ Due to the Heisenberg Uncertainty Principle, we can study a signal with a Fourier
Transform with information in time or with information in frequency BUT NOT BOTH.
▪ E.g. We may know some frequency component occurs often throughout the existence
of a signal but we cannot know the timing this frequency occurs in the signal.
▪The two graphs below demonstrate the difference between examining a signal in the
time domain versus the frequency domain.
ADVANTAGES OVER THE FOURIER TRANSFORM
ADVANTAGES OVER THE FOURIER TRANSFORM
▪ The Wavelet Transform allows us to get both time and scale (the analogue for
frequency for wavelets) information simultaneously.
▪ J represents the “scale” or the “resolution level” at which we would want to examine
the signal.
▪ K represents the “translation” or “shift” in time.
ADVANTAGES OVER THE FOURIER TRANSFORM
▪ The Fourier Transform does NOT represent abrupt changes efficiently
▪ To accurately analyze signals and images with abrupt changes, use the wavelet
transform which is localized in time and frequency
WAVELET ANALYSIS AND TIME SERIES
“Time Series Forecasts Via Wavelets: An Application to Car Sales in the Spanish
Market” by Miguel Ariño, Ph.D.
GENERAL OUTLINE
▪ We will first forecast the time series using conventional time series modeling tools
such as Autoregressive Integrated Moving Average (ARIMA) and Seasonal ARIMA
(SARIMA).
▪ We want to compare the accuracy of these forecasts to forecasts combining wavelet
analysis and ARIMA models.
COMBINING WAVELET ANALYSIS AND ARIMA
▪ Decompose the time series into its long-term trend and seasonal component using
wavelet decomposition and reconstruction.
▪ Using ARIMA models, forecast each component separately.
▪ Combine these forecasted components to get the forecast for the original time series.
DATA
▪ Monthly Car Sales in Spain from January 1974 to December 1994.
▪ 252 total observations
▪ We will use the first 240 observations to build the models and the last 12 to
compare our forecasts with the actual number of sales during 1994.
ARIMA MODELS
▪ There are general rules of thumb to follow when identifying ARIMA models.
▪ Examine the Autocorrelation (correlation between each observation and the past
observations) and Partial Autocorrelation (correlation not accounted for by lags in-
between)
▪ Take differences (if necessary) to remove non-stationarity of the time series
▪ Two good models describe our time series:
▪ ARIMA(0, 1, 1)x(0, 1, 1)12
▪ ARIMA(2, 1, 0)x(0, 1, 1)12
ARIMA (0, 1, 1) X (0, 1, 1)12
▪ While both models give similar forecasts, the first model is simpler so we will use this
one.
▪ No AR component, 1 non-seasonal difference, 1 moving average component
▪ No seasonal AR component, 1 seasonal difference, 1 seasonal moving average
component with seasonal frequency being 12 months
▪ The out-of-sample root mean square error (RMSE) is 16,963.9
WAVELET MODEL – PREPARING TO USE DWT
▪ In order to use the DWT, we need the number of data points to be a power of 2
▪ Center the time series by subtracting its mean 60,603
▪ This centered or zero-mean series is called x = (xt)
▪ Daubechies of order 8 Wavelet basis
WAVELET MODEL - DWT
▪ Apply DWT to our time series x to obtain another series or vector called d which will
represent our scaling and wavelet coefficients.
▪ The DWT can be represented as a matrix, so applying the DWT can be thought of
as:
▪ For each level there will be associated coefficients
LONG-TERM TREND AND SEASONAL COMPONENT
▪ Scalogram is a graph of the amount of “energy” for each level of resolution to
identify the two most dominant resolution levels to use as the long-term trend and
seasonal components.
▪ For each resolution level and amount of shift/translation k
▪ Example for 2nd Resolution level:
SPLITTING THE COEFFICIENTS INTO 2 SETS
▪ There are two major peaks in the scalogram at levels 1 (7th decomposition) and 7
(1st decomposition).
▪ Existence of peaks at high or low levels indicates the existence of high or low
frequency components.
▪ Take coefficients at levels around each of the major peaks and pad each of these 2
coefficient vectors with zeros.
RECONSTRUCTION OF THE COMPONENTS
▪ Now that we have the 2 separate sets of coefficients, let’s reconstruct the individual
components using the IDWT.
BACK TO FORECASTING
▪ We have decomposed our time series into simpler and easier-to-forecast
components.
▪ In order to forecast y we need to add back in the mean of the original time series
that we subtracted before doing the wavelet analysis.
▪ Decomposition and Reconstruction on the boundary
▪ Delete first 36 and last 16 data points of each of our series x, y, and z.
▪ Left with car sales from January 1977 to December 1993
FORECASTING THE COMPONENTS
▪ Seasonality removed from the long-term trend y
▪ Best model for y: ARIMA(1,3,0)
▪ Seasonal model z
▪ Best model for z: ARIMA(0,1,1)12
▪ Forecasts for x
▪ x = y + z
COMPARISON OF THE MODELS
▪ Wavelet Model RMSE = 12,194
▪ SARIMA Model RMSE = 16,964
POTENTIAL APPLICATIONS FOR I KNOW FIRST
▪ Could provide more confidence to investment forecasts
▪ Overlay current forecasting systems on top of each of the decomposed wavelet
components.
▪ Wavelet analysis does not have to be constrained to 1-dimensional problems
▪Could be used for n-dimensional problems which could include considering an entire
portfolio of investments.
USEFUL RESOURCES
▪ Wavelet Toolbox User’s Guide
▪http://web.mit.edu/1.130/WebDocs/wavelet_ug.pdf
▪The Discrete Wavelet Transform in S
▪http://www.stat.ucla.edu/~cocteau/stat204/readings/nasonsilverman.pdf
▪Wavelet Scalograms and Their Applications in Economic Time Series
▪https://www.ime.usp.br/~pam/amv.pdf
▪“Conceptual Wavelets in Digital Signal Processing: An In-Depth, Practical Approach
for the Non-Mathematician” by D. Lee Fugal

Weitere ähnliche Inhalte

Was ist angesagt?

A brief introduction of Artificial neural network by example
A brief introduction of Artificial neural network by exampleA brief introduction of Artificial neural network by example
A brief introduction of Artificial neural network by exampleMrinmoy Majumder
 
presentation on digital signal processing
presentation on digital signal processingpresentation on digital signal processing
presentation on digital signal processingsandhya jois
 
Wavelet Multi-resolution Analysis of High Frequency FX Rates
Wavelet Multi-resolution Analysis of High Frequency FX RatesWavelet Multi-resolution Analysis of High Frequency FX Rates
Wavelet Multi-resolution Analysis of High Frequency FX RatesaiQUANT
 
Introduction to data warehousing
Introduction to data warehousing   Introduction to data warehousing
Introduction to data warehousing Girish Dhareshwar
 
Analog to Digital Conversion
Analog to Digital ConversionAnalog to Digital Conversion
Analog to Digital ConversionSyed Umair
 
Gradient descent method
Gradient descent methodGradient descent method
Gradient descent methodSanghyuk Chun
 
Fast Fourier Transform
Fast Fourier TransformFast Fourier Transform
Fast Fourier Transformop205
 
The Fast Fourier Transform (FFT)
The Fast Fourier Transform (FFT)The Fast Fourier Transform (FFT)
The Fast Fourier Transform (FFT)Oka Danil
 
Coherent systems
Coherent systemsCoherent systems
Coherent systemsCKSunith1
 
Memory mapped I/O and Isolated I/O
Memory mapped I/O and Isolated I/OMemory mapped I/O and Isolated I/O
Memory mapped I/O and Isolated I/OBharat Kharbanda
 
4.2 spatial data mining
4.2 spatial data mining4.2 spatial data mining
4.2 spatial data miningKrish_ver2
 
Introduction to spatial data mining
Introduction to spatial data miningIntroduction to spatial data mining
Introduction to spatial data miningHoang Nguyen
 
DSP_FOEHU - Lec 10 - FIR Filter Design
DSP_FOEHU - Lec 10 - FIR Filter DesignDSP_FOEHU - Lec 10 - FIR Filter Design
DSP_FOEHU - Lec 10 - FIR Filter DesignAmr E. Mohamed
 

Was ist angesagt? (20)

digital filters
digital filtersdigital filters
digital filters
 
Data analysis of weather forecasting
Data analysis of weather forecastingData analysis of weather forecasting
Data analysis of weather forecasting
 
A brief introduction of Artificial neural network by example
A brief introduction of Artificial neural network by exampleA brief introduction of Artificial neural network by example
A brief introduction of Artificial neural network by example
 
Time series forecasting
Time series forecastingTime series forecasting
Time series forecasting
 
presentation on digital signal processing
presentation on digital signal processingpresentation on digital signal processing
presentation on digital signal processing
 
Wavelet Multi-resolution Analysis of High Frequency FX Rates
Wavelet Multi-resolution Analysis of High Frequency FX RatesWavelet Multi-resolution Analysis of High Frequency FX Rates
Wavelet Multi-resolution Analysis of High Frequency FX Rates
 
Introduction to data warehousing
Introduction to data warehousing   Introduction to data warehousing
Introduction to data warehousing
 
Pulse code modulation
Pulse code modulationPulse code modulation
Pulse code modulation
 
Analog to Digital Conversion
Analog to Digital ConversionAnalog to Digital Conversion
Analog to Digital Conversion
 
Gradient descent method
Gradient descent methodGradient descent method
Gradient descent method
 
Fast Fourier Transform
Fast Fourier TransformFast Fourier Transform
Fast Fourier Transform
 
The Fast Fourier Transform (FFT)
The Fast Fourier Transform (FFT)The Fast Fourier Transform (FFT)
The Fast Fourier Transform (FFT)
 
Phase Noise and Jitter Measurements
Phase Noise and Jitter MeasurementsPhase Noise and Jitter Measurements
Phase Noise and Jitter Measurements
 
Coherent systems
Coherent systemsCoherent systems
Coherent systems
 
Memory mapped I/O and Isolated I/O
Memory mapped I/O and Isolated I/OMemory mapped I/O and Isolated I/O
Memory mapped I/O and Isolated I/O
 
4.2 spatial data mining
4.2 spatial data mining4.2 spatial data mining
4.2 spatial data mining
 
Bode plot
Bode plotBode plot
Bode plot
 
Introduction to spatial data mining
Introduction to spatial data miningIntroduction to spatial data mining
Introduction to spatial data mining
 
Dsp presentation
Dsp presentationDsp presentation
Dsp presentation
 
DSP_FOEHU - Lec 10 - FIR Filter Design
DSP_FOEHU - Lec 10 - FIR Filter DesignDSP_FOEHU - Lec 10 - FIR Filter Design
DSP_FOEHU - Lec 10 - FIR Filter Design
 

Ähnlich wie Financial Forecasting Using Wavelet Analysis

Value at Risk Engine
Value at Risk EngineValue at Risk Engine
Value at Risk EngineLov Loothra
 
Brock Butlett Time Series-Great Lakes
Brock Butlett Time Series-Great Lakes Brock Butlett Time Series-Great Lakes
Brock Butlett Time Series-Great Lakes Brock Butlett
 
Different Models Used In Time Series - InsideAIML
Different Models Used In Time Series - InsideAIMLDifferent Models Used In Time Series - InsideAIML
Different Models Used In Time Series - InsideAIMLVijaySharma802
 
Demand time series analysis and forecasting
Demand time series analysis and forecastingDemand time series analysis and forecasting
Demand time series analysis and forecastingM Baddar
 
Cab travel time prediction using ensemble models
Cab travel time prediction using ensemble modelsCab travel time prediction using ensemble models
Cab travel time prediction using ensemble modelsAyan Sengupta
 
Long Memory presentation to SURF
Long Memory presentation to SURFLong Memory presentation to SURF
Long Memory presentation to SURFRichard Hunt
 
Wavelet Based Image Compression Using FPGA
Wavelet Based Image Compression Using FPGAWavelet Based Image Compression Using FPGA
Wavelet Based Image Compression Using FPGADr. Mohieddin Moradi
 
JF608: Quality Control - Unit 3
JF608: Quality Control - Unit 3JF608: Quality Control - Unit 3
JF608: Quality Control - Unit 3Asraf Malik
 
__MESA presentation,Profitable Adaptive Trading Systems.ppt
__MESA presentation,Profitable Adaptive Trading Systems.ppt__MESA presentation,Profitable Adaptive Trading Systems.ppt
__MESA presentation,Profitable Adaptive Trading Systems.pptssuser48f316
 
Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...
Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...
Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...Meir TOLEDANO
 
Anaplan Stat Forecasting Methods.pdf
Anaplan Stat Forecasting Methods.pdfAnaplan Stat Forecasting Methods.pdf
Anaplan Stat Forecasting Methods.pdfVishYrdy
 
Cyclostationary analysis of polytime coded signals for lpi radars
Cyclostationary analysis of polytime coded signals for lpi radarsCyclostationary analysis of polytime coded signals for lpi radars
Cyclostationary analysis of polytime coded signals for lpi radarseSAT Journals
 
Computer aided design of communication systems / Simulation Communication Sys...
Computer aided design of communication systems / Simulation Communication Sys...Computer aided design of communication systems / Simulation Communication Sys...
Computer aided design of communication systems / Simulation Communication Sys...Makan Mohammadi
 
Chapter 6 - Modelling and Control of Converters.pdf
Chapter 6 - Modelling and Control of Converters.pdfChapter 6 - Modelling and Control of Converters.pdf
Chapter 6 - Modelling and Control of Converters.pdfbenson215
 

Ähnlich wie Financial Forecasting Using Wavelet Analysis (20)

Time series analysis
Time series analysisTime series analysis
Time series analysis
 
Value at Risk Engine
Value at Risk EngineValue at Risk Engine
Value at Risk Engine
 
Module 1 (1).pdf
Module 1 (1).pdfModule 1 (1).pdf
Module 1 (1).pdf
 
Brock Butlett Time Series-Great Lakes
Brock Butlett Time Series-Great Lakes Brock Butlett Time Series-Great Lakes
Brock Butlett Time Series-Great Lakes
 
E121 gt2014 26029
E121 gt2014 26029E121 gt2014 26029
E121 gt2014 26029
 
Different Models Used In Time Series - InsideAIML
Different Models Used In Time Series - InsideAIMLDifferent Models Used In Time Series - InsideAIML
Different Models Used In Time Series - InsideAIML
 
Demand time series analysis and forecasting
Demand time series analysis and forecastingDemand time series analysis and forecasting
Demand time series analysis and forecasting
 
Cab travel time prediction using ensemble models
Cab travel time prediction using ensemble modelsCab travel time prediction using ensemble models
Cab travel time prediction using ensemble models
 
Long Memory presentation to SURF
Long Memory presentation to SURFLong Memory presentation to SURF
Long Memory presentation to SURF
 
Wavelet Based Image Compression Using FPGA
Wavelet Based Image Compression Using FPGAWavelet Based Image Compression Using FPGA
Wavelet Based Image Compression Using FPGA
 
Speech technology basics
Speech technology   basicsSpeech technology   basics
Speech technology basics
 
Asp unit 1.pdf
Asp unit 1.pdfAsp unit 1.pdf
Asp unit 1.pdf
 
Daubechies wavelets
Daubechies waveletsDaubechies wavelets
Daubechies wavelets
 
JF608: Quality Control - Unit 3
JF608: Quality Control - Unit 3JF608: Quality Control - Unit 3
JF608: Quality Control - Unit 3
 
__MESA presentation,Profitable Adaptive Trading Systems.ppt
__MESA presentation,Profitable Adaptive Trading Systems.ppt__MESA presentation,Profitable Adaptive Trading Systems.ppt
__MESA presentation,Profitable Adaptive Trading Systems.ppt
 
Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...
Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...
Lightning fast time series modeling and prediction: (S)ARIMA on steroids - ST...
 
Anaplan Stat Forecasting Methods.pdf
Anaplan Stat Forecasting Methods.pdfAnaplan Stat Forecasting Methods.pdf
Anaplan Stat Forecasting Methods.pdf
 
Cyclostationary analysis of polytime coded signals for lpi radars
Cyclostationary analysis of polytime coded signals for lpi radarsCyclostationary analysis of polytime coded signals for lpi radars
Cyclostationary analysis of polytime coded signals for lpi radars
 
Computer aided design of communication systems / Simulation Communication Sys...
Computer aided design of communication systems / Simulation Communication Sys...Computer aided design of communication systems / Simulation Communication Sys...
Computer aided design of communication systems / Simulation Communication Sys...
 
Chapter 6 - Modelling and Control of Converters.pdf
Chapter 6 - Modelling and Control of Converters.pdfChapter 6 - Modelling and Control of Converters.pdf
Chapter 6 - Modelling and Control of Converters.pdf
 

Mehr von I Know First: Daily Market Forecast

Mehr von I Know First: Daily Market Forecast (8)

The Future of Risk Management Part 1: Forecasting Liquidity
The Future of Risk Management Part 1: Forecasting LiquidityThe Future of Risk Management Part 1: Forecasting Liquidity
The Future of Risk Management Part 1: Forecasting Liquidity
 
I Know First Presentation (May 2016)
I Know First Presentation (May 2016)I Know First Presentation (May 2016)
I Know First Presentation (May 2016)
 
Machine Learning, Stock Market and Chaos
Machine Learning, Stock Market and Chaos Machine Learning, Stock Market and Chaos
Machine Learning, Stock Market and Chaos
 
Aktienkurse vorhersagen: Big Data als Lösung
Aktienkurse vorhersagen: Big Data als LösungAktienkurse vorhersagen: Big Data als Lösung
Aktienkurse vorhersagen: Big Data als Lösung
 
Google Stock Forecast: Bullish Algorithmic Forecast For Google Even With Issu...
Google Stock Forecast: Bullish Algorithmic Forecast For Google Even With Issu...Google Stock Forecast: Bullish Algorithmic Forecast For Google Even With Issu...
Google Stock Forecast: Bullish Algorithmic Forecast For Google Even With Issu...
 
Algorithmic Market Outlook: Volatility On The Rise
Algorithmic Market Outlook: Volatility On The RiseAlgorithmic Market Outlook: Volatility On The Rise
Algorithmic Market Outlook: Volatility On The Rise
 
Algorithmic Trading Latest Trends & Developments Lecture By Dr. Lipa Roitman
Algorithmic Trading Latest Trends & Developments Lecture By Dr. Lipa RoitmanAlgorithmic Trading Latest Trends & Developments Lecture By Dr. Lipa Roitman
Algorithmic Trading Latest Trends & Developments Lecture By Dr. Lipa Roitman
 
2014 Market Forecast Based On Algorithms
2014 Market Forecast Based On Algorithms2014 Market Forecast Based On Algorithms
2014 Market Forecast Based On Algorithms
 

Kürzlich hochgeladen

2024 Q1 Crypto Industry Report | CoinGecko
2024 Q1 Crypto Industry Report | CoinGecko2024 Q1 Crypto Industry Report | CoinGecko
2024 Q1 Crypto Industry Report | CoinGeckoCoinGecko
 
Market Morning Updates for 16th April 2024
Market Morning Updates for 16th April 2024Market Morning Updates for 16th April 2024
Market Morning Updates for 16th April 2024Devarsh Vakil
 
NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...
NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...
NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...Amil baba
 
magnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdf
magnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdfmagnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdf
magnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdfHenry Tapper
 
House of Commons ; CDC schemes overview document
House of Commons ; CDC schemes overview documentHouse of Commons ; CDC schemes overview document
House of Commons ; CDC schemes overview documentHenry Tapper
 
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证jdkhjh
 
NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...
NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...
NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...Amil baba
 
Stock Market Brief Deck for "this does not happen often".pdf
Stock Market Brief Deck for "this does not happen often".pdfStock Market Brief Deck for "this does not happen often".pdf
Stock Market Brief Deck for "this does not happen often".pdfMichael Silva
 
(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办
(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办
(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办fqiuho152
 
『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书
『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书
『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书rnrncn29
 
SBP-Market-Operations and market managment
SBP-Market-Operations and market managmentSBP-Market-Operations and market managment
SBP-Market-Operations and market managmentfactical
 
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证rjrjkk
 
Economic Risk Factor Update: April 2024 [SlideShare]
Economic Risk Factor Update: April 2024 [SlideShare]Economic Risk Factor Update: April 2024 [SlideShare]
Economic Risk Factor Update: April 2024 [SlideShare]Commonwealth
 
Tenets of Physiocracy History of Economic
Tenets of Physiocracy History of EconomicTenets of Physiocracy History of Economic
Tenets of Physiocracy History of Economiccinemoviesu
 
Overview of Inkel Unlisted Shares Price.
Overview of Inkel Unlisted Shares Price.Overview of Inkel Unlisted Shares Price.
Overview of Inkel Unlisted Shares Price.Precize Formely Leadoff
 
Economics, Commerce and Trade Management: An International Journal (ECTIJ)
Economics, Commerce and Trade Management: An International Journal (ECTIJ)Economics, Commerce and Trade Management: An International Journal (ECTIJ)
Economics, Commerce and Trade Management: An International Journal (ECTIJ)ECTIJ
 
NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...
NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...
NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...Amil baba
 
Vp Girls near me Delhi Call Now or WhatsApp
Vp Girls near me Delhi Call Now or WhatsAppVp Girls near me Delhi Call Now or WhatsApp
Vp Girls near me Delhi Call Now or WhatsAppmiss dipika
 
212MTAMount Durham University Bachelor's Diploma in Technology
212MTAMount Durham University Bachelor's Diploma in Technology212MTAMount Durham University Bachelor's Diploma in Technology
212MTAMount Durham University Bachelor's Diploma in Technologyz xss
 
The Core Functions of the Bangko Sentral ng Pilipinas
The Core Functions of the Bangko Sentral ng PilipinasThe Core Functions of the Bangko Sentral ng Pilipinas
The Core Functions of the Bangko Sentral ng PilipinasCherylouCamus
 

Kürzlich hochgeladen (20)

2024 Q1 Crypto Industry Report | CoinGecko
2024 Q1 Crypto Industry Report | CoinGecko2024 Q1 Crypto Industry Report | CoinGecko
2024 Q1 Crypto Industry Report | CoinGecko
 
Market Morning Updates for 16th April 2024
Market Morning Updates for 16th April 2024Market Morning Updates for 16th April 2024
Market Morning Updates for 16th April 2024
 
NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...
NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...
NO1 Certified Amil Baba In Lahore Kala Jadu In Lahore Best Amil In Lahore Ami...
 
magnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdf
magnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdfmagnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdf
magnetic-pensions-a-new-blueprint-for-the-dc-landscape.pdf
 
House of Commons ; CDC schemes overview document
House of Commons ; CDC schemes overview documentHouse of Commons ; CDC schemes overview document
House of Commons ; CDC schemes overview document
 
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
 
NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...
NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...
NO1 WorldWide Love marriage specialist baba ji Amil Baba Kala ilam powerful v...
 
Stock Market Brief Deck for "this does not happen often".pdf
Stock Market Brief Deck for "this does not happen often".pdfStock Market Brief Deck for "this does not happen often".pdf
Stock Market Brief Deck for "this does not happen often".pdf
 
(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办
(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办
(办理原版一样)QUT毕业证昆士兰科技大学毕业证学位证留信学历认证成绩单补办
 
『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书
『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书
『澳洲文凭』买科廷大学毕业证书成绩单办理澳洲Curtin文凭学位证书
 
SBP-Market-Operations and market managment
SBP-Market-Operations and market managmentSBP-Market-Operations and market managment
SBP-Market-Operations and market managment
 
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
 
Economic Risk Factor Update: April 2024 [SlideShare]
Economic Risk Factor Update: April 2024 [SlideShare]Economic Risk Factor Update: April 2024 [SlideShare]
Economic Risk Factor Update: April 2024 [SlideShare]
 
Tenets of Physiocracy History of Economic
Tenets of Physiocracy History of EconomicTenets of Physiocracy History of Economic
Tenets of Physiocracy History of Economic
 
Overview of Inkel Unlisted Shares Price.
Overview of Inkel Unlisted Shares Price.Overview of Inkel Unlisted Shares Price.
Overview of Inkel Unlisted Shares Price.
 
Economics, Commerce and Trade Management: An International Journal (ECTIJ)
Economics, Commerce and Trade Management: An International Journal (ECTIJ)Economics, Commerce and Trade Management: An International Journal (ECTIJ)
Economics, Commerce and Trade Management: An International Journal (ECTIJ)
 
NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...
NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...
NO1 Certified Best Amil In Rawalpindi Bangali Baba In Rawalpindi jadu tona ka...
 
Vp Girls near me Delhi Call Now or WhatsApp
Vp Girls near me Delhi Call Now or WhatsAppVp Girls near me Delhi Call Now or WhatsApp
Vp Girls near me Delhi Call Now or WhatsApp
 
212MTAMount Durham University Bachelor's Diploma in Technology
212MTAMount Durham University Bachelor's Diploma in Technology212MTAMount Durham University Bachelor's Diploma in Technology
212MTAMount Durham University Bachelor's Diploma in Technology
 
The Core Functions of the Bangko Sentral ng Pilipinas
The Core Functions of the Bangko Sentral ng PilipinasThe Core Functions of the Bangko Sentral ng Pilipinas
The Core Functions of the Bangko Sentral ng Pilipinas
 

Financial Forecasting Using Wavelet Analysis

  • 1. Wavelet Analysis By Sam Luxenberg 23rd March 2017, Tel Aviv
  • 2. INTRODUCTION TO WAVELETS Applications: ▪ Data Compression ▪ Signal Analysis ▪ Smoothing/De-noising Data ▪ Modeling Abrupt Changes ▪ Pattern Recognition ▪ Solutions to Partial Differential Equations
  • 3. WAVELETS AND FINANCIAL MARKETS ▪Markets are complex chaotic systems that have self-similarity properties. ▪Financial signals can be thought of as fractal signals having self-similarity properties which allows for analysis with wavelets.
  • 4. SIGNAL PROCESSING AND ANALYSIS ▪ The conventional tool is Fourier Analysis which can represent a signal (e.g. audio signal, images, seismic signal, financial signal) as a sum of sinusoids (think cosine and sine curves). ▪ We may be interested in certain cyclic or frequency components present in a signal. ▪ The Fourier Transform allows us to study the signal not just as it relates to time, but also these frequency components.
  • 5. REPRESENTATION OF SIGNALS ▪ In Fourier Analysis, a signal g(t) can be represented as: Where are called the Fourier coefficients (aka Fourier Transform of g(t)) ▪ In Wavelet Analysis, a signal g(t) can be represented as: Where are called the wavelet coefficients of g(t) with respect to the basis
  • 6. SIGNAL PROCESSING AND ANALYSIS ▪ Consider the Discrete Fourier Transform (DFT) Let x0, x1, …, xN-1 be a signal sampled at N points in time. Then the DFT is: where ▪Note that this is a linear transformation of the signal and can therefore be represented as a matrix. ▪ This sum can be thought of as the correlation between the signal xn and the different frequency components.
  • 7. SIGNAL PROCESSING AND ANALYSIS ▪ Consider the Discrete Wavelet Transform (DWT) Let x0, x1, …, x2n-1 be a signal sampled at 2n points in time. Then via the DWT, x has its wavelet decomposition of the form: where is a scaling function. ▪ Note that the DWT is also a linear (and orthogonal) transformation and can therefore be represented as a matrix.
  • 8.
  • 9.
  • 10. GENERAL STRUCTURE OF WAVELET ANALYSIS 2 Steps ▪ Decomposition with the DWT ▪ Reconstruction/Synthesis with the Inverse DWT (IDWT)
  • 11. ONE LEVEL OF WAVELET DECOMPOSITION ▪ The DWT of a signal x is calculated by passing it through a series of filters. ▪ Samples are simultaneously passed through a lower-pass and high-pass filter resulting in a convolution of the two. ▪ (approximation or scaling coefficients) ▪ (detail coefficients)
  • 12.
  • 13. ADVANTAGES OVER FOURIER TRANSFORM ▪ Due to the Heisenberg Uncertainty Principle, we can study a signal with a Fourier Transform with information in time or with information in frequency BUT NOT BOTH. ▪ E.g. We may know some frequency component occurs often throughout the existence of a signal but we cannot know the timing this frequency occurs in the signal. ▪The two graphs below demonstrate the difference between examining a signal in the time domain versus the frequency domain.
  • 14. ADVANTAGES OVER THE FOURIER TRANSFORM
  • 15. ADVANTAGES OVER THE FOURIER TRANSFORM ▪ The Wavelet Transform allows us to get both time and scale (the analogue for frequency for wavelets) information simultaneously. ▪ J represents the “scale” or the “resolution level” at which we would want to examine the signal. ▪ K represents the “translation” or “shift” in time.
  • 16. ADVANTAGES OVER THE FOURIER TRANSFORM ▪ The Fourier Transform does NOT represent abrupt changes efficiently ▪ To accurately analyze signals and images with abrupt changes, use the wavelet transform which is localized in time and frequency
  • 17. WAVELET ANALYSIS AND TIME SERIES “Time Series Forecasts Via Wavelets: An Application to Car Sales in the Spanish Market” by Miguel Ariño, Ph.D.
  • 18. GENERAL OUTLINE ▪ We will first forecast the time series using conventional time series modeling tools such as Autoregressive Integrated Moving Average (ARIMA) and Seasonal ARIMA (SARIMA). ▪ We want to compare the accuracy of these forecasts to forecasts combining wavelet analysis and ARIMA models.
  • 19. COMBINING WAVELET ANALYSIS AND ARIMA ▪ Decompose the time series into its long-term trend and seasonal component using wavelet decomposition and reconstruction. ▪ Using ARIMA models, forecast each component separately. ▪ Combine these forecasted components to get the forecast for the original time series.
  • 20. DATA ▪ Monthly Car Sales in Spain from January 1974 to December 1994. ▪ 252 total observations ▪ We will use the first 240 observations to build the models and the last 12 to compare our forecasts with the actual number of sales during 1994.
  • 21.
  • 22. ARIMA MODELS ▪ There are general rules of thumb to follow when identifying ARIMA models. ▪ Examine the Autocorrelation (correlation between each observation and the past observations) and Partial Autocorrelation (correlation not accounted for by lags in- between) ▪ Take differences (if necessary) to remove non-stationarity of the time series ▪ Two good models describe our time series: ▪ ARIMA(0, 1, 1)x(0, 1, 1)12 ▪ ARIMA(2, 1, 0)x(0, 1, 1)12
  • 23. ARIMA (0, 1, 1) X (0, 1, 1)12 ▪ While both models give similar forecasts, the first model is simpler so we will use this one. ▪ No AR component, 1 non-seasonal difference, 1 moving average component ▪ No seasonal AR component, 1 seasonal difference, 1 seasonal moving average component with seasonal frequency being 12 months ▪ The out-of-sample root mean square error (RMSE) is 16,963.9
  • 24.
  • 25. WAVELET MODEL – PREPARING TO USE DWT ▪ In order to use the DWT, we need the number of data points to be a power of 2 ▪ Center the time series by subtracting its mean 60,603 ▪ This centered or zero-mean series is called x = (xt) ▪ Daubechies of order 8 Wavelet basis
  • 26. WAVELET MODEL - DWT ▪ Apply DWT to our time series x to obtain another series or vector called d which will represent our scaling and wavelet coefficients. ▪ The DWT can be represented as a matrix, so applying the DWT can be thought of as: ▪ For each level there will be associated coefficients
  • 27.
  • 28. LONG-TERM TREND AND SEASONAL COMPONENT ▪ Scalogram is a graph of the amount of “energy” for each level of resolution to identify the two most dominant resolution levels to use as the long-term trend and seasonal components. ▪ For each resolution level and amount of shift/translation k ▪ Example for 2nd Resolution level:
  • 29.
  • 30. SPLITTING THE COEFFICIENTS INTO 2 SETS ▪ There are two major peaks in the scalogram at levels 1 (7th decomposition) and 7 (1st decomposition). ▪ Existence of peaks at high or low levels indicates the existence of high or low frequency components. ▪ Take coefficients at levels around each of the major peaks and pad each of these 2 coefficient vectors with zeros.
  • 31. RECONSTRUCTION OF THE COMPONENTS ▪ Now that we have the 2 separate sets of coefficients, let’s reconstruct the individual components using the IDWT.
  • 32. BACK TO FORECASTING ▪ We have decomposed our time series into simpler and easier-to-forecast components. ▪ In order to forecast y we need to add back in the mean of the original time series that we subtracted before doing the wavelet analysis. ▪ Decomposition and Reconstruction on the boundary ▪ Delete first 36 and last 16 data points of each of our series x, y, and z. ▪ Left with car sales from January 1977 to December 1993
  • 33.
  • 34. FORECASTING THE COMPONENTS ▪ Seasonality removed from the long-term trend y ▪ Best model for y: ARIMA(1,3,0) ▪ Seasonal model z ▪ Best model for z: ARIMA(0,1,1)12 ▪ Forecasts for x ▪ x = y + z
  • 35.
  • 36. COMPARISON OF THE MODELS ▪ Wavelet Model RMSE = 12,194 ▪ SARIMA Model RMSE = 16,964
  • 37. POTENTIAL APPLICATIONS FOR I KNOW FIRST ▪ Could provide more confidence to investment forecasts ▪ Overlay current forecasting systems on top of each of the decomposed wavelet components. ▪ Wavelet analysis does not have to be constrained to 1-dimensional problems ▪Could be used for n-dimensional problems which could include considering an entire portfolio of investments.
  • 38. USEFUL RESOURCES ▪ Wavelet Toolbox User’s Guide ▪http://web.mit.edu/1.130/WebDocs/wavelet_ug.pdf ▪The Discrete Wavelet Transform in S ▪http://www.stat.ucla.edu/~cocteau/stat204/readings/nasonsilverman.pdf ▪Wavelet Scalograms and Their Applications in Economic Time Series ▪https://www.ime.usp.br/~pam/amv.pdf ▪“Conceptual Wavelets in Digital Signal Processing: An In-Depth, Practical Approach for the Non-Mathematician” by D. Lee Fugal