SlideShare une entreprise Scribd logo
1  sur  92
Télécharger pour lire hors ligne
Predicting Time Periods of Excessive Price Volatility:
The Case of Rice
Dr. Ramon Clarete Alfonso Labao
University of the Philippines, School of Economics
September 25, 2013
Flow of Presentation
Overview
Methodology
Empirical Results
Conclusion and Constraints
Overview:
Overview:
Importance of being forewarned of extreme food price-volatility:
Overview:
Importance of being forewarned of extreme food price-volatility:
provides time to undertake cooperation
prevent herding and self-fulfilling crises
avoid repetition of 2008 rice crisis
prevent welfare costs for the poor
Overview:
Importance of being forewarned of extreme food price-volatility:
provides time to undertake cooperation
prevent herding and self-fulfilling crises
avoid repetition of 2008 rice crisis
prevent welfare costs for the poor
G20 report stresses importance of accurate and timely market information
Our Methodology:
Our Methodology:
- Use daily rice future prices
(5407 days from Sep 1991 to Mar 2013)
Our Methodology:
- Use daily rice future prices
(5407 days from Sep 1991 to Mar 2013)
- Identify time-periods of excessive price volatility
(via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology:
Two-step spline-backfitted-kernel estimation / GPD distribution)
Our Methodology:
- Use daily rice future prices
(5407 days from Sep 1991 to Mar 2013)
- Identify time-periods of excessive price volatility
(via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology:
Two-step spline-backfitted-kernel estimation / GPD distribution)
- Develop Early Warning Signals
(via Particle Swarm Optimization (PSO))
First Two Parts:
- Use daily rice future prices
(5407 days from Sep 1991 to Mar 2013)
- Identify time-periods of excessive price volatility
(via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology:
Two-step spline-backfitted-kernel estimation / GPD distribution)
- Develop Early Warning Signals
(via Particle Swarm Optimization (PSO))
Actual Prices:
Snapshot of actual prices:
Converted into Price Returns:
Converted into Price Returns:
Converted into Price Returns:
where price returns refer to day-to-day price movements, or:
Converted into Price Returns:
where price returns refer to day-to-day price movements, or:
price return = ln(
price of today
price of yesterday
)
Create Thresholds via MTY’s two-step methodology:
Create Thresholds via MTY’s two-step methodology:
Create Thresholds via MTY’s two-step methodology:
Methodology:
MTY’s Two-step method:
Methodology:
MTY’s Two-step method:
Construct a nonparametric trend via spline-backfitted-kernel:
rt = m0 +
d
a=1
ma(Xta) + (h0 +
d
a=1
ha(Xta)1/2
) t
Estimate the high-order 95 percent quantile via Generalized Pareto Distribution (GPD):
ˆqt(α) = ˆk+1 +
ˆβt
ˆγt
((
(1 − α)
(k/N)
)−ˆγt
− 1)
Methodology:
MTY’s Two-step method:
Construct a nonparametric trend via spline-backfitted-kernel:
rt = m0 +
d
a=1
ma(Xta) + (h0 +
d
a=1
ha(Xta)1/2
) t
Estimate the high-order 95 percent quantile via Generalized Pareto Distribution (GPD):
ˆqt(α) = ˆk+1 +
ˆβt
ˆγt
((
(1 − α)
(k/N)
)−ˆγt
− 1)
Estimated 95 percent Conditional Quantiles:
ˆqt(α/rt−1, rt−2) = ˜m(rt−1, rt−2) + [(˜h(rt−1, rt−2))1/2
ˆqt(α)]
How are Time Periods of Excessive Price Volatility (EPV) defined?
How are Time Periods of Excessive Price Volatility (EPV) defined?
Martins-Filho, Maximo Torero and Feng Yao’s definition of time periods of EPV:
EPV Definition:
Time-Periods whereby the preceding 60 days experienced a significantly high amount
of extreme positive price returns over the 95 percent conditional quantile.
How are Time Periods of Excessive Price Volatility (EPV) defined?
Martins-Filho, Maximo Torero and Feng Yao’s definition of time periods of EPV:
EPV Definition:
Time-Periods whereby the preceding 60 days experienced a significantly high amount
of extreme positive price returns over the 95 percent conditional quantile.
... at least 7 instances of extreme positive price returns within a span of 60 days
1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Oct 1994) EPV cluster:
1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Oct 1994) EPV cluster:
3rd (Mar 1999 - Sep 1999) and 4th (Jun 2001 - Aug 2001) EPV cluster:
5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV cluster:
5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV cluster:
7th (Mar 2008 - Nov 2008) and 8th (Apr 2009 - May 2009) EPV cluster:
9th (Jul 2011) EPV cluster:
In Summary:
In Summary:
501 days of excessive price-volatility (EPV)
Nine (9) clusters of EPV
In Summary:
501 days of excessive price-volatility (EPV)
Nine (9) clusters of EPV
In Summary:
501 days of excessive price-volatility (EPV)
Nine (9) clusters of EPV
... Of the above clusters of EPV, only four (4) are severe
Third Part:
- Use daily rice future prices
(5407 days from Sep 1991 to Mar 2013)
- Identify time-periods of excessive price volatility
(via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology:
Two-step spline-backfitted-kernel estimation / GPD distribution)
- Develop Early Warning Signals
(via Particle Swarm Optimization (PSO))
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
We set the mirror-image’s lag at 60 days prior to an EPV cluster...
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Next Task: Looking for a Good Early Warning Signal
Parameters of the Early Warning Signals:
Next Task: Looking for a Good Early Warning Signal
Parameters of the Early Warning Signals:
Window of Observation
Next Task: Looking for a Good Early Warning Signal
Parameters of the Early Warning Signals:
Window of Observation
Lower-Order Quantile Level
Next Task: Looking for a Good Early Warning Signal
Parameters of the Early Warning Signals:
Window of Observation
Lower-Order Quantile Level
Frequency of Breach
Next Task: Looking for a Good Early Warning Signal
Parameters of the Early Warning Signals:
Window of Observation
Lower-Order Quantile Level
Frequency of Breach
Scope
Next Task: Looking for a Good Early Warning Signal
Parameters of the Early Warning Signals:
Window of Observation
Lower-Order Quantile Level
Frequency of Breach
Scope
Together, these parameters generate a time-based variable, namely...
Next Task: Looking for a Good Early Warning Signal
... the scope’s time-coverage
What makes a good early warning signal:
What makes a good early warning signal:
accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms
What makes a good early warning signal:
accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms
good lead time: there’s reasonable lead time before EPV cluster
What makes a good early warning signal:
accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms
good lead time: there’s reasonable lead time before EPV cluster
comprehensive: signals pre-empt almost all EPV clusters
What makes a good early warning signal:
accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms
good lead time: there’s reasonable lead time before EPV cluster
comprehensive: signals pre-empt almost all EPV clusters
Basically...
We create a new set of trends (using spbk) and solve for parameters that meet above
requirements
We expressed the criteria (accurate, comprehensive, good lead time), into a single
additive objective function that can be minimized:
Objective Function for Minimization:
f (X) = [(
count1
laghvp
) ∗ 1.35] + [
count2
hvp
] + [(accuracy) ∗ 1.35]+
[(
totalews
totaldays
) ∗ 1.25] + [
totalscope
totaldays
]
We expressed the criteria (accurate, comprehensive, good lead time), into a single
additive objective function that can be minimized:
Objective Function for Minimization:
f (X) = [(
count1
laghvp
) ∗ 1.35] + [
count2
hvp
] + [(accuracy) ∗ 1.35]+
[(
totalews
totaldays
) ∗ 1.25] + [
totalscope
totaldays
]
Where:
count1: total no. of lagged mirror-images of time periods of EPV not covered by
the scopes of the signals
laghvp: total no. of lagged mirror-images of time periods of EPV
count2: total no. of actual time periods of EPV not covered by the scopes’
time-coverage of the early warning signals
hvp: total no. of actual time periods of EPV
accuracy: predictive accuracy of the early warning signal with following ratio:
[1-((no.of high.vol.periods captured by scope’s time-coverage) / (hvp)].
total ews: total no. of early warning signals
total scope: total no. of days covered by the scopes’ time-coverage of the early
warning signals
total days: total no. of days within the sample timeframe (1991 to 2013): 5407
future days
We expressed the criteria (accurate, comprehensive, good lead time), into a single
additive objective function that can be minimized:
Objective Function for Minimization:
f (X) = [(
count1
laghvp
) ∗ 1.35] + [
count2
hvp
] + [(accuracy) ∗ 1.35]+
[(
totalews
totaldays
) ∗ 1.25] + [
totalscope
totaldays
]
We expressed the criteria (accurate, comprehensive, good lead time), into a single
additive objective function that can be minimized:
Objective Function for Minimization:
f (X) = [(
count1
laghvp
) ∗ 1.35] + [
count2
hvp
] + [(accuracy) ∗ 1.35]+
[(
totalews
totaldays
) ∗ 1.25] + [
totalscope
totaldays
]
First two (2) terms - optimize good lead-time and comprehensiveness
We expressed the criteria (accurate, comprehensive, good lead time), into a single
additive objective function that can be minimized:
Objective Function for Minimization:
f (X) = [(
count1
laghvp
) ∗ 1.35] + [
count2
hvp
] + [(accuracy) ∗ 1.35]+
[(
totalews
totaldays
) ∗ 1.25] + [
totalscope
totaldays
]
First two (2) terms - optimize good lead-time and comprehensiveness
Last three (3) terms - minimize false alarms and improve accuracy.
We use R’s PSO package to minimize the objective function...
We use R’s PSO package to minimize the objective function...
PSO: Particle Swarm Optimization
PSO is a meta-heuristic designed for functions that are hard to optimize using
traditional optimization procedures (due to several peaks, non-linearities, long-forms,
etc..)
We use R’s PSO package to minimize the objective function...
PSO: Particle Swarm Optimization
PSO is a meta-heuristic designed for functions that are hard to optimize using
traditional optimization procedures (due to several peaks, non-linearities, long-forms,
etc..)
PSO Parameter Search Range:
We use R’s PSO package to minimize the objective function...
PSO: Particle Swarm Optimization
PSO is a meta-heuristic designed for functions that are hard to optimize using
traditional optimization procedures (due to several peaks, non-linearities, long-forms,
etc..)
PSO Parameter Search Range:
Window of Observation: from 1 to 40 future days
Quantile Level: from 50 percent to 99 percent quantile
Frequency of Breach: from 1 to 5
Scope of Early Warning Signal: from 60 to 250 future days
After 1000 iterations, we obtain the following optimal solution:
After 1000 iterations, we obtain the following optimal solution:
Optimal Solution:
Window of Observation: 11.61 future days
Quantile Level: 91.31 percent
Frequency of Breach: 4.18
Scope of Early Warning Signal: 107.06 future days
After 1000 iterations, we obtain the following optimal solution:
Optimal Solution:
Window of Observation: 11.61 future days
Quantile Level: 91.31 percent
Frequency of Breach: 4.18
Scope of Early Warning Signal: 107.06 future days
Given these parameters, an early warning signal will be activated...
After 1000 iterations, we obtain the following optimal solution:
Optimal Solution:
Window of Observation: 11.61 future days
Quantile Level: 91.31 percent
Frequency of Breach: 4.18
Scope of Early Warning Signal: 107.06 future days
Given these parameters, an early warning signal will be activated...
... once the 91.31 percent conditional quantile is breached five (5) times within
11.61 future days..
Just a review...
Just a review...
What makes a good early warning signal:
accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms
good lead time: there’s reasonable lead time before EPV cluster
comprehensive: signals pre-empt almost all EPV clusters
We obtain these Performance Statistics:
We obtain these Performance Statistics:
accuracy of ews: 94.44 percent predictive power
comprehensiveness: 99.40 percent of time periods of EPV are covered
lead-time before EPV cluster: average of 43 days from the earliest signal to start of
high-volatility time-period
Below is a summary of the different EPV clusters and the lead-times of the ews:
Below is a summary of the different EPV clusters and the lead-times of the ews:
Total of 125 early warning signals activated from Sep 1991 to Mar 2013.
Below is a summary of the different EPV clusters and the lead-times of the ews:
Total of 125 early warning signals activated from Sep 1991 to Mar 2013.
This is only 2.3 percent of the time.
Below is a summary of the different EPV clusters and the lead-times of the ews:
Average of 43 days between earliest ews and start of an EPV cluster
Below is a summary of the different EPV clusters and the lead-times of the ews:
Average of 43 days between earliest ews and start of an EPV cluster
But among the four (4) severe EPV clusters, there is an average lead-time of 61
days
Below is a summary of the different EPV clusters and the lead-times of the ews:
Average of 43 days between earliest ews and start of an EPV cluster
But among the four (4) severe EPV clusters, there is an average lead-time of 61
days
The unflagged 2011 EPV cluster is not severe..
Below is a summary of the different EPV clusters and the lead-times of the ews:
Average of 43 days between earliest ews and start of an EPV cluster
But among the four (4) severe EPV clusters, there is an average lead-time of 61
days
The unflagged 2011 EPV cluster is not severe..
All EPV are sufficiently covered from beginning till end by the scopes of the
ews..
for illustration...
1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Sep 1994) EPV Clusters:
1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Sep 1994) EPV Clusters:
3rd EPV Cluster (Mar 1999 - Sep 1999)and 4th (Jun 2001 - Aug 2001) EPV Clusters:
5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV Clusters:
5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV Clusters:
7th (Mar 2008 - Nov 2008) and 8th (Apr 2009 - May 2009) EPV Clusters:
5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV Clusters:
7th (Mar 2008 - Nov 2008) and 8th (Apr 2009 - May 2009) EPV Clusters:
9th (Jul 2011) EPV Cluster:
Constraints:
Constraints:
usage of daily future prices, not daily spot prices
sensitivity analysis
application to other commodities
Thank You.

Contenu connexe

Similaire à Predicting Time Periiods of Excessive Price Volatility: The Case of Rice- Ramon Clarete and Alfonso Labao

Application of Regression and Neural Network Models in Computing Forecasts fo...
Application of Regression and Neural Network Models in Computing Forecasts fo...Application of Regression and Neural Network Models in Computing Forecasts fo...
Application of Regression and Neural Network Models in Computing Forecasts fo...
iosrjce
 
CH&Cie white paper value-at-risk in tuburlent times_VaR
CH&Cie white paper value-at-risk in tuburlent times_VaRCH&Cie white paper value-at-risk in tuburlent times_VaR
CH&Cie white paper value-at-risk in tuburlent times_VaR
Thibault Le Pomellec
 
Grouped time-series forecasting: Application to regional infant mortality counts
Grouped time-series forecasting: Application to regional infant mortality countsGrouped time-series forecasting: Application to regional infant mortality counts
Grouped time-series forecasting: Application to regional infant mortality counts
hanshang
 
Forecasting
ForecastingForecasting
Forecasting
3abooodi
 

Similaire à Predicting Time Periiods of Excessive Price Volatility: The Case of Rice- Ramon Clarete and Alfonso Labao (20)

THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEMTHE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
THE STUDY OF CUCKOO OPTIMIZATION ALGORITHM FOR PRODUCTION PLANNING PROBLEM
 
Application of Regression and Neural Network Models in Computing Forecasts fo...
Application of Regression and Neural Network Models in Computing Forecasts fo...Application of Regression and Neural Network Models in Computing Forecasts fo...
Application of Regression and Neural Network Models in Computing Forecasts fo...
 
Design of Kalman filter for Airborne Applications
Design of Kalman filter for Airborne ApplicationsDesign of Kalman filter for Airborne Applications
Design of Kalman filter for Airborne Applications
 
presentation data fusion methods ex.pptx
presentation data fusion methods ex.pptxpresentation data fusion methods ex.pptx
presentation data fusion methods ex.pptx
 
Modeling adoptions and the stages of the diffusion of innovations
Modeling adoptions and the stages of the diffusion of innovationsModeling adoptions and the stages of the diffusion of innovations
Modeling adoptions and the stages of the diffusion of innovations
 
Wearable Accelerometer Optimal Positions for Human Motion Recognition(LifeTec...
Wearable Accelerometer Optimal Positions for Human Motion Recognition(LifeTec...Wearable Accelerometer Optimal Positions for Human Motion Recognition(LifeTec...
Wearable Accelerometer Optimal Positions for Human Motion Recognition(LifeTec...
 
IFTA2020 Kei Nakagawa
IFTA2020 Kei NakagawaIFTA2020 Kei Nakagawa
IFTA2020 Kei Nakagawa
 
forecasting methods
forecasting methodsforecasting methods
forecasting methods
 
Airline revenue maximization through genetic algorithm
Airline revenue maximization through genetic algorithmAirline revenue maximization through genetic algorithm
Airline revenue maximization through genetic algorithm
 
CH&Cie white paper value-at-risk in tuburlent times_VaR
CH&Cie white paper value-at-risk in tuburlent times_VaRCH&Cie white paper value-at-risk in tuburlent times_VaR
CH&Cie white paper value-at-risk in tuburlent times_VaR
 
Oriskany' TCS Multi-Factor Timing - Factor Investing White Paper
Oriskany' TCS Multi-Factor Timing - Factor Investing White PaperOriskany' TCS Multi-Factor Timing - Factor Investing White Paper
Oriskany' TCS Multi-Factor Timing - Factor Investing White Paper
 
Grouped time-series forecasting: Application to regional infant mortality counts
Grouped time-series forecasting: Application to regional infant mortality countsGrouped time-series forecasting: Application to regional infant mortality counts
Grouped time-series forecasting: Application to regional infant mortality counts
 
Stochastic Vol Forecasting
Stochastic Vol ForecastingStochastic Vol Forecasting
Stochastic Vol Forecasting
 
Forecasting
ForecastingForecasting
Forecasting
 

Plus de International Food Policy Research Institute

Plus de International Food Policy Research Institute (20)

The History of Cooking Oil Fortification in Indonesia: Government Support for...
The History of Cooking Oil Fortification in Indonesia: Government Support for...The History of Cooking Oil Fortification in Indonesia: Government Support for...
The History of Cooking Oil Fortification in Indonesia: Government Support for...
 
Food Fortification Policies in the Asia Region
Food Fortification Policies in the Asia RegionFood Fortification Policies in the Asia Region
Food Fortification Policies in the Asia Region
 
Wheat and Cooking Oil/Ghee Fortification in Pakistan
Wheat and Cooking Oil/Ghee Fortification in PakistanWheat and Cooking Oil/Ghee Fortification in Pakistan
Wheat and Cooking Oil/Ghee Fortification in Pakistan
 
China's Food Safety regulatory system: Achievements, Challenges and Suggestions
China's Food Safety regulatory system: Achievements, Challenges and SuggestionsChina's Food Safety regulatory system: Achievements, Challenges and Suggestions
China's Food Safety regulatory system: Achievements, Challenges and Suggestions
 
Institutional and Governance Innovation in Thailand’s Food System: The Role o...
Institutional and Governance Innovation in Thailand’s Food System: The Role o...Institutional and Governance Innovation in Thailand’s Food System: The Role o...
Institutional and Governance Innovation in Thailand’s Food System: The Role o...
 
Compliance of Producers and Adoption of Consumers in the Case of Food Safety ...
Compliance of Producers and Adoption of Consumers in the Case of Food Safety ...Compliance of Producers and Adoption of Consumers in the Case of Food Safety ...
Compliance of Producers and Adoption of Consumers in the Case of Food Safety ...
 
Current Status of Agricultural Biotechnology in Thailand
Current Status of Agricultural Biotechnology in ThailandCurrent Status of Agricultural Biotechnology in Thailand
Current Status of Agricultural Biotechnology in Thailand
 
Creating and Implementing Biosafety Regulations: The Philippine Experience
Creating and Implementing Biosafety Regulations: The Philippine Experience Creating and Implementing Biosafety Regulations: The Philippine Experience
Creating and Implementing Biosafety Regulations: The Philippine Experience
 
Making Vegetable Markets Work
Making Vegetable Markets WorkMaking Vegetable Markets Work
Making Vegetable Markets Work
 
Soybean Value Chains for Rural Development
Soybean Value Chains for Rural DevelopmentSoybean Value Chains for Rural Development
Soybean Value Chains for Rural Development
 
Findings from the Study on Nutrition-Sensitive Value Chains in the Feed the F...
Findings from the Study on Nutrition-Sensitive Value Chains in the Feed the F...Findings from the Study on Nutrition-Sensitive Value Chains in the Feed the F...
Findings from the Study on Nutrition-Sensitive Value Chains in the Feed the F...
 
Farm Production, Market Access and Dietary Diversity in China’s Poor Rural Ho...
Farm Production, Market Access and Dietary Diversity in China’s Poor Rural Ho...Farm Production, Market Access and Dietary Diversity in China’s Poor Rural Ho...
Farm Production, Market Access and Dietary Diversity in China’s Poor Rural Ho...
 
The Livestock Sector in India: Progress and Challenges
The Livestock Sector in India: Progress and ChallengesThe Livestock Sector in India: Progress and Challenges
The Livestock Sector in India: Progress and Challenges
 
Production Diversity and Market Access for Predicting Animal-source Food Cons...
Production Diversity and Market Access for Predicting Animal-source Food Cons...Production Diversity and Market Access for Predicting Animal-source Food Cons...
Production Diversity and Market Access for Predicting Animal-source Food Cons...
 
The Quiet Revolution in Myanmar’s Aquaculture Value Chain
The Quiet Revolution in Myanmar’s Aquaculture Value ChainThe Quiet Revolution in Myanmar’s Aquaculture Value Chain
The Quiet Revolution in Myanmar’s Aquaculture Value Chain
 
Does e-commerce Increase Food Consumption in Rural Areas? Evidence from China
Does e-commerce Increase Food Consumption in Rural Areas? Evidence from ChinaDoes e-commerce Increase Food Consumption in Rural Areas? Evidence from China
Does e-commerce Increase Food Consumption in Rural Areas? Evidence from China
 
Impacting at Scale: From .5% to + 40%
Impacting at Scale: From .5% to + 40% Impacting at Scale: From .5% to + 40%
Impacting at Scale: From .5% to + 40%
 
Regulatory Cooperation in ASEAN Good Agricultural Practices
Regulatory Cooperation in ASEAN Good Agricultural PracticesRegulatory Cooperation in ASEAN Good Agricultural Practices
Regulatory Cooperation in ASEAN Good Agricultural Practices
 
Traditional Versus Modern Milk Marketing Chains in India: Implications for Sm...
Traditional Versus Modern Milk Marketing Chains in India: Implications for Sm...Traditional Versus Modern Milk Marketing Chains in India: Implications for Sm...
Traditional Versus Modern Milk Marketing Chains in India: Implications for Sm...
 
Pakistan’s Multi-Sectoral Nutrition Strategy
Pakistan’s Multi-Sectoral Nutrition Strategy Pakistan’s Multi-Sectoral Nutrition Strategy
Pakistan’s Multi-Sectoral Nutrition Strategy
 

Dernier

Architecting Cloud Native Applications
Architecting Cloud Native ApplicationsArchitecting Cloud Native Applications
Architecting Cloud Native Applications
WSO2
 

Dernier (20)

Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
 
Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...
 
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingRepurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
 
Data Cloud, More than a CDP by Matt Robison
Data Cloud, More than a CDP by Matt RobisonData Cloud, More than a CDP by Matt Robison
Data Cloud, More than a CDP by Matt Robison
 
Architecting Cloud Native Applications
Architecting Cloud Native ApplicationsArchitecting Cloud Native Applications
Architecting Cloud Native Applications
 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a Fresher
 
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot TakeoffStrategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
 
Automating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps ScriptAutomating Google Workspace (GWS) & more with Apps Script
Automating Google Workspace (GWS) & more with Apps Script
 
A Beginners Guide to Building a RAG App Using Open Source Milvus
A Beginners Guide to Building a RAG App Using Open Source MilvusA Beginners Guide to Building a RAG App Using Open Source Milvus
A Beginners Guide to Building a RAG App Using Open Source Milvus
 
DBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor PresentationDBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor Presentation
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdf
 
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ..."I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptx
 
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
 
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
 
GenAI Risks & Security Meetup 01052024.pdf
GenAI Risks & Security Meetup 01052024.pdfGenAI Risks & Security Meetup 01052024.pdf
GenAI Risks & Security Meetup 01052024.pdf
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024
 

Predicting Time Periiods of Excessive Price Volatility: The Case of Rice- Ramon Clarete and Alfonso Labao

  • 1. Predicting Time Periods of Excessive Price Volatility: The Case of Rice Dr. Ramon Clarete Alfonso Labao University of the Philippines, School of Economics September 25, 2013
  • 2. Flow of Presentation Overview Methodology Empirical Results Conclusion and Constraints
  • 4. Overview: Importance of being forewarned of extreme food price-volatility:
  • 5. Overview: Importance of being forewarned of extreme food price-volatility: provides time to undertake cooperation prevent herding and self-fulfilling crises avoid repetition of 2008 rice crisis prevent welfare costs for the poor
  • 6. Overview: Importance of being forewarned of extreme food price-volatility: provides time to undertake cooperation prevent herding and self-fulfilling crises avoid repetition of 2008 rice crisis prevent welfare costs for the poor G20 report stresses importance of accurate and timely market information
  • 8. Our Methodology: - Use daily rice future prices (5407 days from Sep 1991 to Mar 2013)
  • 9. Our Methodology: - Use daily rice future prices (5407 days from Sep 1991 to Mar 2013) - Identify time-periods of excessive price volatility (via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology: Two-step spline-backfitted-kernel estimation / GPD distribution)
  • 10. Our Methodology: - Use daily rice future prices (5407 days from Sep 1991 to Mar 2013) - Identify time-periods of excessive price volatility (via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology: Two-step spline-backfitted-kernel estimation / GPD distribution) - Develop Early Warning Signals (via Particle Swarm Optimization (PSO))
  • 11. First Two Parts: - Use daily rice future prices (5407 days from Sep 1991 to Mar 2013) - Identify time-periods of excessive price volatility (via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology: Two-step spline-backfitted-kernel estimation / GPD distribution) - Develop Early Warning Signals (via Particle Swarm Optimization (PSO))
  • 16. Converted into Price Returns: where price returns refer to day-to-day price movements, or:
  • 17. Converted into Price Returns: where price returns refer to day-to-day price movements, or: price return = ln( price of today price of yesterday )
  • 18. Create Thresholds via MTY’s two-step methodology:
  • 19. Create Thresholds via MTY’s two-step methodology:
  • 20. Create Thresholds via MTY’s two-step methodology:
  • 22. Methodology: MTY’s Two-step method: Construct a nonparametric trend via spline-backfitted-kernel: rt = m0 + d a=1 ma(Xta) + (h0 + d a=1 ha(Xta)1/2 ) t Estimate the high-order 95 percent quantile via Generalized Pareto Distribution (GPD): ˆqt(α) = ˆk+1 + ˆβt ˆγt (( (1 − α) (k/N) )−ˆγt − 1)
  • 23. Methodology: MTY’s Two-step method: Construct a nonparametric trend via spline-backfitted-kernel: rt = m0 + d a=1 ma(Xta) + (h0 + d a=1 ha(Xta)1/2 ) t Estimate the high-order 95 percent quantile via Generalized Pareto Distribution (GPD): ˆqt(α) = ˆk+1 + ˆβt ˆγt (( (1 − α) (k/N) )−ˆγt − 1) Estimated 95 percent Conditional Quantiles: ˆqt(α/rt−1, rt−2) = ˜m(rt−1, rt−2) + [(˜h(rt−1, rt−2))1/2 ˆqt(α)]
  • 24. How are Time Periods of Excessive Price Volatility (EPV) defined?
  • 25. How are Time Periods of Excessive Price Volatility (EPV) defined? Martins-Filho, Maximo Torero and Feng Yao’s definition of time periods of EPV: EPV Definition: Time-Periods whereby the preceding 60 days experienced a significantly high amount of extreme positive price returns over the 95 percent conditional quantile.
  • 26. How are Time Periods of Excessive Price Volatility (EPV) defined? Martins-Filho, Maximo Torero and Feng Yao’s definition of time periods of EPV: EPV Definition: Time-Periods whereby the preceding 60 days experienced a significantly high amount of extreme positive price returns over the 95 percent conditional quantile. ... at least 7 instances of extreme positive price returns within a span of 60 days
  • 27. 1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Oct 1994) EPV cluster:
  • 28. 1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Oct 1994) EPV cluster: 3rd (Mar 1999 - Sep 1999) and 4th (Jun 2001 - Aug 2001) EPV cluster:
  • 29. 5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV cluster:
  • 30. 5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV cluster: 7th (Mar 2008 - Nov 2008) and 8th (Apr 2009 - May 2009) EPV cluster:
  • 31. 9th (Jul 2011) EPV cluster:
  • 33. In Summary: 501 days of excessive price-volatility (EPV) Nine (9) clusters of EPV
  • 34. In Summary: 501 days of excessive price-volatility (EPV) Nine (9) clusters of EPV
  • 35. In Summary: 501 days of excessive price-volatility (EPV) Nine (9) clusters of EPV ... Of the above clusters of EPV, only four (4) are severe
  • 36. Third Part: - Use daily rice future prices (5407 days from Sep 1991 to Mar 2013) - Identify time-periods of excessive price volatility (via Martins-Filho, Maximo Torero, and Feng Yao’s (IFPRI, 2010) methodology: Two-step spline-backfitted-kernel estimation / GPD distribution) - Develop Early Warning Signals (via Particle Swarm Optimization (PSO))
  • 37. Next Task: Looking for a Good Early Warning Signal
  • 38. Next Task: Looking for a Good Early Warning Signal
  • 39. Next Task: Looking for a Good Early Warning Signal
  • 40. Next Task: Looking for a Good Early Warning Signal
  • 41. Next Task: Looking for a Good Early Warning Signal
  • 42. Next Task: Looking for a Good Early Warning Signal
  • 43. Next Task: Looking for a Good Early Warning Signal
  • 44. Next Task: Looking for a Good Early Warning Signal We set the mirror-image’s lag at 60 days prior to an EPV cluster...
  • 45. Next Task: Looking for a Good Early Warning Signal
  • 46. Next Task: Looking for a Good Early Warning Signal
  • 47. Next Task: Looking for a Good Early Warning Signal Parameters of the Early Warning Signals:
  • 48. Next Task: Looking for a Good Early Warning Signal Parameters of the Early Warning Signals: Window of Observation
  • 49. Next Task: Looking for a Good Early Warning Signal Parameters of the Early Warning Signals: Window of Observation Lower-Order Quantile Level
  • 50. Next Task: Looking for a Good Early Warning Signal Parameters of the Early Warning Signals: Window of Observation Lower-Order Quantile Level Frequency of Breach
  • 51. Next Task: Looking for a Good Early Warning Signal Parameters of the Early Warning Signals: Window of Observation Lower-Order Quantile Level Frequency of Breach Scope
  • 52. Next Task: Looking for a Good Early Warning Signal Parameters of the Early Warning Signals: Window of Observation Lower-Order Quantile Level Frequency of Breach Scope Together, these parameters generate a time-based variable, namely...
  • 53. Next Task: Looking for a Good Early Warning Signal ... the scope’s time-coverage
  • 54. What makes a good early warning signal:
  • 55. What makes a good early warning signal: accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms
  • 56. What makes a good early warning signal: accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms good lead time: there’s reasonable lead time before EPV cluster
  • 57. What makes a good early warning signal: accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms good lead time: there’s reasonable lead time before EPV cluster comprehensive: signals pre-empt almost all EPV clusters
  • 58. What makes a good early warning signal: accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms good lead time: there’s reasonable lead time before EPV cluster comprehensive: signals pre-empt almost all EPV clusters Basically... We create a new set of trends (using spbk) and solve for parameters that meet above requirements
  • 59. We expressed the criteria (accurate, comprehensive, good lead time), into a single additive objective function that can be minimized: Objective Function for Minimization: f (X) = [( count1 laghvp ) ∗ 1.35] + [ count2 hvp ] + [(accuracy) ∗ 1.35]+ [( totalews totaldays ) ∗ 1.25] + [ totalscope totaldays ]
  • 60. We expressed the criteria (accurate, comprehensive, good lead time), into a single additive objective function that can be minimized: Objective Function for Minimization: f (X) = [( count1 laghvp ) ∗ 1.35] + [ count2 hvp ] + [(accuracy) ∗ 1.35]+ [( totalews totaldays ) ∗ 1.25] + [ totalscope totaldays ] Where: count1: total no. of lagged mirror-images of time periods of EPV not covered by the scopes of the signals laghvp: total no. of lagged mirror-images of time periods of EPV count2: total no. of actual time periods of EPV not covered by the scopes’ time-coverage of the early warning signals hvp: total no. of actual time periods of EPV accuracy: predictive accuracy of the early warning signal with following ratio: [1-((no.of high.vol.periods captured by scope’s time-coverage) / (hvp)]. total ews: total no. of early warning signals total scope: total no. of days covered by the scopes’ time-coverage of the early warning signals total days: total no. of days within the sample timeframe (1991 to 2013): 5407 future days
  • 61. We expressed the criteria (accurate, comprehensive, good lead time), into a single additive objective function that can be minimized: Objective Function for Minimization: f (X) = [( count1 laghvp ) ∗ 1.35] + [ count2 hvp ] + [(accuracy) ∗ 1.35]+ [( totalews totaldays ) ∗ 1.25] + [ totalscope totaldays ]
  • 62. We expressed the criteria (accurate, comprehensive, good lead time), into a single additive objective function that can be minimized: Objective Function for Minimization: f (X) = [( count1 laghvp ) ∗ 1.35] + [ count2 hvp ] + [(accuracy) ∗ 1.35]+ [( totalews totaldays ) ∗ 1.25] + [ totalscope totaldays ] First two (2) terms - optimize good lead-time and comprehensiveness
  • 63. We expressed the criteria (accurate, comprehensive, good lead time), into a single additive objective function that can be minimized: Objective Function for Minimization: f (X) = [( count1 laghvp ) ∗ 1.35] + [ count2 hvp ] + [(accuracy) ∗ 1.35]+ [( totalews totaldays ) ∗ 1.25] + [ totalscope totaldays ] First two (2) terms - optimize good lead-time and comprehensiveness Last three (3) terms - minimize false alarms and improve accuracy.
  • 64. We use R’s PSO package to minimize the objective function...
  • 65. We use R’s PSO package to minimize the objective function... PSO: Particle Swarm Optimization PSO is a meta-heuristic designed for functions that are hard to optimize using traditional optimization procedures (due to several peaks, non-linearities, long-forms, etc..)
  • 66. We use R’s PSO package to minimize the objective function... PSO: Particle Swarm Optimization PSO is a meta-heuristic designed for functions that are hard to optimize using traditional optimization procedures (due to several peaks, non-linearities, long-forms, etc..) PSO Parameter Search Range:
  • 67. We use R’s PSO package to minimize the objective function... PSO: Particle Swarm Optimization PSO is a meta-heuristic designed for functions that are hard to optimize using traditional optimization procedures (due to several peaks, non-linearities, long-forms, etc..) PSO Parameter Search Range: Window of Observation: from 1 to 40 future days Quantile Level: from 50 percent to 99 percent quantile Frequency of Breach: from 1 to 5 Scope of Early Warning Signal: from 60 to 250 future days
  • 68. After 1000 iterations, we obtain the following optimal solution:
  • 69. After 1000 iterations, we obtain the following optimal solution: Optimal Solution: Window of Observation: 11.61 future days Quantile Level: 91.31 percent Frequency of Breach: 4.18 Scope of Early Warning Signal: 107.06 future days
  • 70. After 1000 iterations, we obtain the following optimal solution: Optimal Solution: Window of Observation: 11.61 future days Quantile Level: 91.31 percent Frequency of Breach: 4.18 Scope of Early Warning Signal: 107.06 future days Given these parameters, an early warning signal will be activated...
  • 71. After 1000 iterations, we obtain the following optimal solution: Optimal Solution: Window of Observation: 11.61 future days Quantile Level: 91.31 percent Frequency of Breach: 4.18 Scope of Early Warning Signal: 107.06 future days Given these parameters, an early warning signal will be activated... ... once the 91.31 percent conditional quantile is breached five (5) times within 11.61 future days..
  • 73. Just a review... What makes a good early warning signal: accurate: scope’s time-coverage covers an EPV cluster, minimal false alarms good lead time: there’s reasonable lead time before EPV cluster comprehensive: signals pre-empt almost all EPV clusters
  • 74. We obtain these Performance Statistics:
  • 75. We obtain these Performance Statistics: accuracy of ews: 94.44 percent predictive power comprehensiveness: 99.40 percent of time periods of EPV are covered lead-time before EPV cluster: average of 43 days from the earliest signal to start of high-volatility time-period
  • 76. Below is a summary of the different EPV clusters and the lead-times of the ews:
  • 77. Below is a summary of the different EPV clusters and the lead-times of the ews: Total of 125 early warning signals activated from Sep 1991 to Mar 2013.
  • 78. Below is a summary of the different EPV clusters and the lead-times of the ews: Total of 125 early warning signals activated from Sep 1991 to Mar 2013. This is only 2.3 percent of the time.
  • 79. Below is a summary of the different EPV clusters and the lead-times of the ews: Average of 43 days between earliest ews and start of an EPV cluster
  • 80. Below is a summary of the different EPV clusters and the lead-times of the ews: Average of 43 days between earliest ews and start of an EPV cluster But among the four (4) severe EPV clusters, there is an average lead-time of 61 days
  • 81. Below is a summary of the different EPV clusters and the lead-times of the ews: Average of 43 days between earliest ews and start of an EPV cluster But among the four (4) severe EPV clusters, there is an average lead-time of 61 days The unflagged 2011 EPV cluster is not severe..
  • 82. Below is a summary of the different EPV clusters and the lead-times of the ews: Average of 43 days between earliest ews and start of an EPV cluster But among the four (4) severe EPV clusters, there is an average lead-time of 61 days The unflagged 2011 EPV cluster is not severe.. All EPV are sufficiently covered from beginning till end by the scopes of the ews..
  • 84. 1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Sep 1994) EPV Clusters:
  • 85. 1st (Jul 1993 - Feb 1994) and 2nd (Jul 1994 - Sep 1994) EPV Clusters: 3rd EPV Cluster (Mar 1999 - Sep 1999)and 4th (Jun 2001 - Aug 2001) EPV Clusters:
  • 86. 5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV Clusters:
  • 87. 5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV Clusters: 7th (Mar 2008 - Nov 2008) and 8th (Apr 2009 - May 2009) EPV Clusters:
  • 88. 5th (Jul 2002 - Aug 2002) and 6th (May 2003 - Oct 2003) EPV Clusters: 7th (Mar 2008 - Nov 2008) and 8th (Apr 2009 - May 2009) EPV Clusters:
  • 89. 9th (Jul 2011) EPV Cluster:
  • 91. Constraints: usage of daily future prices, not daily spot prices sensitivity analysis application to other commodities