Rafael Serrano profesor de la Universidad del Rosario
Resumen:
In the first part of the talk, I will present an introduction to stochastic affine short rate models for term structure of yield curves In the second part, I will focus on a recursive affine cascade with persistent factors for which the number of parameters, under specifications, is invariant to the size of the state space and converges to a stochastic limit as the number of factors goes to infinity. The cascade construction thereby overcomes dimensionality difficulties associated with general affine models. We contrast two specfifications of the model using linear Kalman filter for a panel of Colombian sovereign yields.
"Correlated Volatility Shocks" by Dr. Xiao Qiao, Researcher at SummerHaven In...Quantopian
Commonality in idiosyncratic volatility cannot be completely explained by time-varying volatility. After removing the effects of time-varying volatility, idiosyncratic volatility innovations are still positively correlated. This result suggests correlated volatility shocks contribute to the comovement in idiosyncratic volatility.
Motivated by this fact, we propose the Dynamic Factor Correlation (DFC) model, which fits the data well and captures the cross-sectional correlations in idiosyncratic volatility innovations. We decompose the common factor in idiosyncratic volatility (CIV) of Herskovic et al. (2016) into the volatility innovation factor (VIN) and time-varying volatility factor (TVV). Whereas VIN is associated with strong variation in average returns, TVV is only weakly priced in the cross section
A strategy that takes a long position in the portfolio with the lowest VIN and TVV betas, and a short position in the portfolio with the highest VIN and TVV betas earns average returns of 8.0% per year.
Hedging, Arbitrage, and Optimality with Superlinear Frictionsguasoni
In a continuous-time model with multiple assets described by cadlag processes, this paper characterizes superhedging prices, absence of arbitrage, and utility maximizing strategies, under general frictions that make execution prices arbitrarily unfavorable for high trading intensity. With such frictions, dual elements correspond to a pair of a shadow execution price combined with an equivalent martingale measure. For utility functions defined on the real line, optimal strategies exist even if arbitrage is present, because it is not scalable at will.
Nonlinear Price Impact and Portfolio Choiceguasoni
In a market with price-impact proportional to a power of the order flow, we derive optimal trading policies and their implied welfare for long-term investors with constant relative risk aversion, who trade one safe asset and one risky asset that follows geometric Brownian motion. These quantities admit asymptotic explicit formulas up to a structural constant that depends only on the price-impact exponent. Trading rates are finite as with linear impact, but they are lower near the target portfolio, and higher away from the target. The model nests the square-root impact law and, as extreme cases, linear impact and proportional transaction costs.
1 factor vs.2 factor gaussian model for zero coupon bond pricing finalFinancial Algorithms
Financial Algorithms describes the comparison between and relevance of Gaussian one and two factor models in today's interest rate environments across US, European and Asian markets. Negative short rates seem to be the new norm of interest rate markets, especially in Euro-zone & somewhat in US , where poor demand and very low inflation dragging down interest rates in a negative zone. One and Two Factor Gaussian Models under Hull-White Setup can accommodate such scenarios and address the cases of curve steeping of longer end of the zero curve wherein short rates hover in negative zones.
"Correlated Volatility Shocks" by Dr. Xiao Qiao, Researcher at SummerHaven In...Quantopian
Commonality in idiosyncratic volatility cannot be completely explained by time-varying volatility. After removing the effects of time-varying volatility, idiosyncratic volatility innovations are still positively correlated. This result suggests correlated volatility shocks contribute to the comovement in idiosyncratic volatility.
Motivated by this fact, we propose the Dynamic Factor Correlation (DFC) model, which fits the data well and captures the cross-sectional correlations in idiosyncratic volatility innovations. We decompose the common factor in idiosyncratic volatility (CIV) of Herskovic et al. (2016) into the volatility innovation factor (VIN) and time-varying volatility factor (TVV). Whereas VIN is associated with strong variation in average returns, TVV is only weakly priced in the cross section
A strategy that takes a long position in the portfolio with the lowest VIN and TVV betas, and a short position in the portfolio with the highest VIN and TVV betas earns average returns of 8.0% per year.
Hedging, Arbitrage, and Optimality with Superlinear Frictionsguasoni
In a continuous-time model with multiple assets described by cadlag processes, this paper characterizes superhedging prices, absence of arbitrage, and utility maximizing strategies, under general frictions that make execution prices arbitrarily unfavorable for high trading intensity. With such frictions, dual elements correspond to a pair of a shadow execution price combined with an equivalent martingale measure. For utility functions defined on the real line, optimal strategies exist even if arbitrage is present, because it is not scalable at will.
Nonlinear Price Impact and Portfolio Choiceguasoni
In a market with price-impact proportional to a power of the order flow, we derive optimal trading policies and their implied welfare for long-term investors with constant relative risk aversion, who trade one safe asset and one risky asset that follows geometric Brownian motion. These quantities admit asymptotic explicit formulas up to a structural constant that depends only on the price-impact exponent. Trading rates are finite as with linear impact, but they are lower near the target portfolio, and higher away from the target. The model nests the square-root impact law and, as extreme cases, linear impact and proportional transaction costs.
1 factor vs.2 factor gaussian model for zero coupon bond pricing finalFinancial Algorithms
Financial Algorithms describes the comparison between and relevance of Gaussian one and two factor models in today's interest rate environments across US, European and Asian markets. Negative short rates seem to be the new norm of interest rate markets, especially in Euro-zone & somewhat in US , where poor demand and very low inflation dragging down interest rates in a negative zone. One and Two Factor Gaussian Models under Hull-White Setup can accommodate such scenarios and address the cases of curve steeping of longer end of the zero curve wherein short rates hover in negative zones.
This paper analyzes the swap rates issued by the China Inter-bank Offered Rate(CHIBOR) and
selects the one-year FR007 daily data from January 1st, 2019 to June 30th, 2019 as a sample. To fit the data,
we conduct Monte Carlo simulation with several typical continuous short-term swap rate models such as the
Merton model, the Vasicek model, the CIR model, etc. These models contain both linear forms and nonlinear
forms and each has both drift terms and diffusion terms. After empirical analysis, we obtain the parameter
values in Euler-Maruyama scheme and relevant statistical characteristics of each model. The results show that
most of the short-term swap rate models can fit the swap rates and reflect the change of trend, while the CKLSO
model performs best.
Transition matrices and PD’s term structure - Anna CornagliaLászló Árvai
A transition matrix is a square matrix describing the probabilities of moving from one state to another in a dynamic system. In each row there are the probabilities of moving, from the state represented by that row, to the other states. Thus each row of a transition matrix adds to one.
Se presenta la estimación de la curva cero cupón, o vector de precios, para Costa Rica usando metaheurísticas de optimización combinatoria, muy superior al método clásico de Newton. Se usan datos reales para Costa Rica.
In this work, we propose to apply trust region optimization to deep reinforcement
learning using a recently proposed Kronecker-factored approximation to
the curvature. We extend the framework of natural policy gradient and propose
to optimize both the actor and the critic using Kronecker-factored approximate
curvature (K-FAC) with trust region; hence we call our method Actor Critic using
Kronecker-Factored Trust Region (ACKTR). To the best of our knowledge, this
is the first scalable trust region natural gradient method for actor-critic methods.
It is also a method that learns non-trivial tasks in continuous control as well as
discrete control policies directly from raw pixel inputs. We tested our approach
across discrete domains in Atari games as well as continuous domains in the MuJoCo
environment. With the proposed methods, we are able to achieve higher
rewards and a 2- to 3-fold improvement in sample efficiency on average, compared
to previous state-of-the-art on-policy actor-critic methods. Code is available at
https://github.com/openai/baselines.
Research on the Trading Strategy Based On Interest Rate Term Structure Change...inventionjournals
Bond pricing errors exist in the bond market universally, the formation of the reasons for its formation has been controversial. In this paper, in order to obtain the pricing error, the authors first estimate the term structure of interest rate of China's interbank market by using the three spline model and the Svensson model. Then, the author using the moving average model and time series model to build the bond trading strategy based on the pricing error. Through the simulation of our bond portfolio trading, the result shows that bond trading can obtain about 11 basis points of the annual excess return based on bond pricing errors, and the excess return rate is not caused by different bond liquidity or risk characteristics, instead is due to the effective economic information included in the bond pricing error.
Even tho Pi network is not listed on any exchange yet.
Buying/Selling or investing in pi network coins is highly possible through the help of vendors. You can buy from vendors[ buy directly from the pi network miners and resell it]. I will leave the telegram contact of my personal vendor.
@Pi_vendor_247
Abhay Bhutada Leads Poonawalla Fincorp To Record Low NPA And Unprecedented Gr...Vighnesh Shashtri
Under the leadership of Abhay Bhutada, Poonawalla Fincorp has achieved record-low Non-Performing Assets (NPA) and witnessed unprecedented growth. Bhutada's strategic vision and effective management have significantly enhanced the company's financial health, showcasing a robust performance in the financial sector. This achievement underscores the company's resilience and ability to thrive in a competitive market, setting a new benchmark for operational excellence in the industry.
how to sell pi coins in South Korea profitably.DOT TECH
Yes. You can sell your pi network coins in South Korea or any other country, by finding a verified pi merchant
What is a verified pi merchant?
Since pi network is not launched yet on any exchange, the only way you can sell pi coins is by selling to a verified pi merchant, and this is because pi network is not launched yet on any exchange and no pre-sale or ico offerings Is done on pi.
Since there is no pre-sale, the only way exchanges can get pi is by buying from miners. So a pi merchant facilitates these transactions by acting as a bridge for both transactions.
How can i find a pi vendor/merchant?
Well for those who haven't traded with a pi merchant or who don't already have one. I will leave the telegram id of my personal pi merchant who i trade pi with.
Tele gram: @Pi_vendor_247
#pi #sell #nigeria #pinetwork #picoins #sellpi #Nigerian #tradepi #pinetworkcoins #sellmypi
how can I sell pi coins after successfully completing KYCDOT TECH
Pi coins is not launched yet in any exchange 💱 this means it's not swappable, the current pi displaying on coin market cap is the iou version of pi. And you can learn all about that on my previous post.
RIGHT NOW THE ONLY WAY you can sell pi coins is through verified pi merchants. A pi merchant is someone who buys pi coins and resell them to exchanges and crypto whales. Looking forward to hold massive quantities of pi coins before the mainnet launch.
This is because pi network is not doing any pre-sale or ico offerings, the only way to get my coins is from buying from miners. So a merchant facilitates the transactions between the miners and these exchanges holding pi.
I and my friends has sold more than 6000 pi coins successfully with this method. I will be happy to share the contact of my personal pi merchant. The one i trade with, if you have your own merchant you can trade with them. For those who are new.
Message: @Pi_vendor_247 on telegram.
I wouldn't advise you selling all percentage of the pi coins. Leave at least a before so its a win win during open mainnet. Have a nice day pioneers ♥️
#kyc #mainnet #picoins #pi #sellpi #piwallet
#pinetwork
how to sell pi coins effectively (from 50 - 100k pi)DOT TECH
Anywhere in the world, including Africa, America, and Europe, you can sell Pi Network Coins online and receive cash through online payment options.
Pi has not yet been launched on any exchange because we are currently using the confined Mainnet. The planned launch date for Pi is June 28, 2026.
Reselling to investors who want to hold until the mainnet launch in 2026 is currently the sole way to sell.
Consequently, right now. All you need to do is select the right pi network provider.
Who is a pi merchant?
An individual who buys coins from miners on the pi network and resells them to investors hoping to hang onto them until the mainnet is launched is known as a pi merchant.
debuts.
I'll provide you the Telegram username
@Pi_vendor_247
The secret way to sell pi coins effortlessly.DOT TECH
Well as we all know pi isn't launched yet. But you can still sell your pi coins effortlessly because some whales in China are interested in holding massive pi coins. And they are willing to pay good money for it. If you are interested in selling I will leave a contact for you. Just telegram this number below. I sold about 3000 pi coins to him and he paid me immediately.
Telegram: @Pi_vendor_247
What website can I sell pi coins securely.DOT TECH
Currently there are no website or exchange that allow buying or selling of pi coins..
But you can still easily sell pi coins, by reselling it to exchanges/crypto whales interested in holding thousands of pi coins before the mainnet launch.
Who is a pi merchant?
A pi merchant is someone who buys pi coins from miners and resell to these crypto whales and holders of pi..
This is because pi network is not doing any pre-sale. The only way exchanges can get pi is by buying from miners and pi merchants stands in between the miners and the exchanges.
How can I sell my pi coins?
Selling pi coins is really easy, but first you need to migrate to mainnet wallet before you can do that. I will leave the telegram contact of my personal pi merchant to trade with.
Tele-gram.
@Pi_vendor_247
Yes of course, you can easily start mining pi network coin today and sell to legit pi vendors in the United States.
Here the telegram contact of my personal vendor.
@Pi_vendor_247
#pi network #pi coins #legit #passive income
#US
If you are looking for a pi coin investor. Then look no further because I have the right one he is a pi vendor (he buy and resell to whales in China). I met him on a crypto conference and ever since I and my friends have sold more than 10k pi coins to him And he bought all and still want more. I will drop his telegram handle below just send him a message.
@Pi_vendor_247
where can I find a legit pi merchant onlineDOT TECH
Yes. This is very easy what you need is a recommendation from someone who has successfully traded pi coins before with a merchant.
Who is a pi merchant?
A pi merchant is someone who buys pi network coins and resell them to Investors looking forward to hold thousands of pi coins before the open mainnet.
I will leave the telegram contact of my personal pi merchant to trade with
@Pi_vendor_247
Currently pi network is not tradable on binance or any other exchange because we are still in the enclosed mainnet.
Right now the only way to sell pi coins is by trading with a verified merchant.
What is a pi merchant?
A pi merchant is someone verified by pi network team and allowed to barter pi coins for goods and services.
Since pi network is not doing any pre-sale The only way exchanges like binance/huobi or crypto whales can get pi is by buying from miners. And a merchant stands in between the exchanges and the miners.
I will leave the telegram contact of my personal pi merchant. I and my friends has traded more than 6000pi coins successfully
Tele-gram
@Pi_vendor_247
Turin Startup Ecosystem 2024 - Ricerca sulle Startup e il Sistema dell'Innov...Quotidiano Piemontese
Turin Startup Ecosystem 2024
Una ricerca de il Club degli Investitori, in collaborazione con ToTeM Torino Tech Map e con il supporto della ESCP Business School e di Growth Capital
The Role of Non-Banking Financial Companies (NBFCs)
Affine cascade models for term structure dynamics of sovereign yield curves
1. Sigma Variant Cascade Model for Term
Structure of Interest Rates in Colombia
Laura Natalia Lopez (Precia)
Rafael Serrano (URosario)
Sep 20th - 2019
1
2. Contents
Introduction
Motivation
Literature Review
A Cascade Model for the Interest Rate
Multifrequency Cascade
Factors Representation
Bond Prices and Risk Premia
Bond Prices and Constant Risk Premia
Dimension-Invariant Term Structures
Base specification
Data and Estimation
Empirical Results
Conclusions
2
4. Term structure of interest rates and the yield curve
Yield (to maturity) of a fixed-income instrument or debt contract is the
internal rate of return (IRR) on cash flows of the contract.
The term structure (TS) of interest rates is the mathematical relation-
ship between yields across different terms or maturities (2 month, 2
year, 20 year, etc. ...) for similar debt contracts.
The yield curve (YC) is the plot or graphical representation of the TS
of certain debt contracts in a given currency.
4
6. Why is the YC important?
• Pricing and valuation of other financial instruments and cash
flows
• Benchmark for determining lending and savings rates in the
credit market
• Relationship to business cycle and economic growth.
Indeed, the slope of the yield curve (e.g. difference between 10-
year Treasury bond rate and the 3-month Treasury) is one of the
most powerful predictors of future economic growth, inflation, and
recessions.
A positively sloped yield curve is often an indicator of inflationary
growth. An inverted yield curve is often an indicator of recession.
6
8. What model?
In order to model the TS, we need to choose first the quantity that
serves as state variable and can help determine the dynamics of TS
from observed yields.
8
9. What model?
In order to model the TS, we need to choose first the quantity that
serves as state variable and can help determine the dynamics of TS
from observed yields.
We choose the instantaneous short (-term) rate.
8
10. What model?
In order to model the TS, we need to choose first the quantity that
serves as state variable and can help determine the dynamics of TS
from observed yields.
We choose the instantaneous short (-term) rate.
For its definition, we assume there exists a continuum of zero-coupon
(ZC) bonds, one for each positive term-to-maturity (TTM) τ > 0
• Time measured in years
• P(t, τ) = price at time t of a ZC bond with TTM τ > 0
• y(t, τ) = yield-to-maturity (YTM) defined as the (continuously
compounded, annualized) IRR of the ZC bond
P(t, τ) = e−y(t,τ)τ
8
11. What model?
In order to model the TS, we need to choose first the quantity that
serves as state variable and can help determine the dynamics of TS
from observed yields.
We choose the instantaneous short (-term) rate.
For its definition, we assume there exists a continuum of zero-coupon
(ZC) bonds, one for each positive term-to-maturity (TTM) τ > 0
• Time measured in years
• P(t, τ) = price at time t of a ZC bond with TTM τ > 0
• y(t, τ) = yield-to-maturity (YTM) defined as the (continuously
compounded, annualized) IRR of the ZC bond
P(t, τ) = e−y(t,τ)τ
• Short rate rt := lim∆t→0 y(t, ∆t)
8
12. Why short rate?
• 1st generation of stochastic interest rate models use the
instantaneous short rate as the state variable.
• Two key advantages: very simple yet highly tractable, as they
often lead to analytic formulae for bonds and other associated
financial instruments.
• Drawback: instantaneous short rate is a mathematical idealisation
rather than something that can be observed directly in the market.
In the past decade, short-rate models have to some extent, been
superseded by the LIBOR or SWAP market models in which the
stochastic state variable is a set of benchmark forward LIBOR or
SWAP rates.
Nonetheless, short-rate models remain still very popular among
both academic and practitioners.
9
13. Contribution
• We compare two short-rate models of recursive cascade with
heterogeneously persistent factors that revert in mean to the
immediate previous factor.
• Under certain specifications, the number of parameters is
invariant to the number of cascade factors.
• These models are adjusted to determine the dynamics of the
term structure of interest rates in Colombia.
• The standard model has an assumption in the volatility of the
model that always makes it constant.
• The sigma variant model (SV), proposed in this work, relaxes this
assumption by giving it a structure dependent on the size of the
cascade.
10
14. Literature
• Arbitrage free TSM’s (Duffie and Kan (1996), affine
specifications; Dai and Singlenton (2000, 2002))
• Low-dimensional DTSM’s, limitations of a model with only one
factor (Duan and Simonato (1999), Bolder (2001), Zeytun and
Gupta (2007))
• DTSM’s have problems in dimensionality and parameter
identification (Duffee and Stanton (2008); Duffe (2011))
11
15. Literature
• The 2-factor models have different conclusions (Nowman (2010)
vs. Walther (2003)). While Chatterjee (2005) concludes that a
three-factor model is adequate.
• Calvet, Laurent and Fisher (2018) propose a multifactorial
cascade model that under certain assumptions maintains its
dimension invariant.
• In Colombia there have been no studies beyond 3 factors.
Rodriguez (2016) concludes that 1-factor Vasicek model has a
poor forecast fit and Velsquez et al (2016) 3-factor model fits well
the curve.
12
16. Introduction
A Cascade Model for the Interest Rate
Multifrequency Cascade
Factors Representation
Bond Prices and Risk Premia
Bond Prices and Constant Risk Premia
Dimension-Invariant Term Structures
Data and Estimation
Empirical Results
Conclusions
13
17. Affine TS model
A short rate model is said to provide an affine TS if the prices of the
ZC bonds have the form
P(t, τ) = exp[−b(τ)T
Xt − c(τ)]
with Xt a multivariate stochastic process that models macro- and mi-
croeconomic unobservable variables that affect the yield curve across
terms (short, medium and long)
14
18. Cascade affine TS model
Model assumptions
• Time is continuous with an infinite horizon t ∈ [0, ∞).
• Probability Space {Ω, F, P, (Ft )t 0}.
• P is the measure of physical probability.
• Wt = (W1,t , ..., Wn,t )T
n- dimensional Wiener process or
Brownian motion with independent components
We propose
• Dynamics of short-rate is modeled on the P measure by a
cascade of n mean-reverting difusions
• Long-term rate θr ∈ R is constant
• Adjustment speeds κ speed at which factors return to their
long-term trend
15
19. Multifrequency Cascade
The dynamics of the n− dimensional state vector Xt = (x1,t , . . . , xn,t )
T
is defined as
Assumption 1 (Factors Structure)
x0,t is the long-term rate equal to θr constant, at which the process
reverts in mean. The diffusion cascade is defined for the factors
dxj,t = κj (xj−1,t − xj,t )dt + σj dWj,t (1)
for all j ∈ {1, ..., n}, with κ1, ...κn, σ1, ..., σn strictly positive constants.
Adjustment speeds are strictly increasing κ1 < κ2 < ... < κn.
16
20. Multifrequency Cascade
The dynamics of the n− dimensional state vector Xt = (x1,t , . . . , xn,t )
T
is defined as
Assumption 1 (Factors Structure)
x0,t is the long-term rate equal to θr constant, at which the process
reverts in mean. The diffusion cascade is defined for the factors
dxj,t = κj (xj−1,t − xj,t )dt + σj dWj,t (1)
for all j ∈ {1, ..., n}, with κ1, ...κn, σ1, ..., σn strictly positive constants.
Adjustment speeds are strictly increasing κ1 < κ2 < ... < κn.
• The first factor x1,t is an Ornstein-Uhlenbeck process with mean
reversion to θr .
• For each j ≥ 1 the jth
factor xj,t reverts in mean towards the
(j − 1)th
factor xj−1,t .
• The factors are less persistent when j increases
16
21. Assumption 2 (Short Rate)
The instantaneous interest rate coincides with the last factor: rt = xn,t
for all t 0.
17
22. Assumption 2 (Short Rate)
The instantaneous interest rate coincides with the last factor: rt = xn,t
for all t 0.
the states vector Xt = (x1,t , ..., xn,t )T
∈ Rn
satisfies:
dXt = κ(θ − Xt )dt + Σ1/2
dWt , (2)
Where θ = (θ, ..., θ)T
, Σ is a diagonal matrix with constant elements
σ2
t , ..., σ2
n, and κ is a matrix with diagonal elements κj,j = κj and ele-
ments outside the diagonal κj,j−1 = −κj .
17
23. Σ =
σ2
1 0 0 . . . 0
0 σ2
2 0 . . . 0
0 0 σ2
3 . . . 0
...
...
...
... 0
0 0 0 0 σ2
n
κ =
κ1 0 0 . . . 0
−κ2 κ2 0 . . . 0
0 −κ3 κ3 . . . 0
...
...
...
... 0
0 0 0 −κn κn
Each equation separates the factors according to their average rever-
sion rates, that is, how quickly, in the face of a shock, they return to
their steady or long-term state.
18
24. Representation of Short Rate Factors
Proposition (Factors representation for the short rate)
The instantaneous interest rate rt is a affine function of the initial
state vector X0 and of integrals of the innovation driving each factor
between 0 and t.
rt = θr +
n
j=1
aj (t)(xj,0 − θr ) +
n
j=1
σj
ˆ t
0
aj (t − s)dWj,s (3)
19
25. Representation of Short Rate Factors
Proposition (Factors representation for the short rate)
The instantaneous interest rate rt is a affine function of the initial
state vector X0 and of integrals of the innovation driving each factor
between 0 and t.
rt = θr +
n
j=1
aj (t)(xj,0 − θr ) +
n
j=1
σj
ˆ t
0
aj (t − s)dWj,s (3)
The long-term expectation of the short rate limt→inf EP
t (rt+τ ), is θr .
19
27. Response Functions
Each function aj (τ) is the product convolution of the exponential prob-
ability density functions
Proposition (Response Functions)
For all j < n the response functions aj (τ) are hump-shaped and their
maximum response horizons are monotonically decreasing with j.
The function satisfies the closed-form expression
aj (τ) =
n
i=j
αi,j κi e−κi τ
, with αi,j =
κj+1 . . . κn
κi
n
k=j,k=i (κk − κi )
(4)
20
28. Response Functions
Each function aj (τ) is the product convolution of the exponential prob-
ability density functions
Proposition (Response Functions)
For all j < n the response functions aj (τ) are hump-shaped and their
maximum response horizons are monotonically decreasing with j.
The function satisfies the closed-form expression
aj (τ) =
n
i=j
αi,j κi e−κi τ
, with αi,j =
κj+1 . . . κn
κi
n
k=j,k=i (κk − κi )
(4)
Moreover
0 ≤
n
j=1
aj (τ) ≤ 1
for all τ ≥ 0. Akkouchi (2018).
20
29. Risk Premia and Bond Prices
Suppose there are no arbitrage opportunities, then there exists a risk-
neutral measure pricing measure Q
Then the value in time t of a zero coupon bond, with a monetary unit
as par value and maturity t + τ is given by
P(Xt , τ) = EQ
t exp(−
ˆ t+τ
t
rsds)
21
30. Risk Premia and Bond Prices
Suppose there are no arbitrage opportunities, then there exists a risk-
neutral measure pricing measure Q
Then the value in time t of a zero coupon bond, with a monetary unit
as par value and maturity t + τ is given by
P(Xt , τ) = EQ
t exp(−
ˆ t+τ
t
rsds)
Let γ denote the risk premia vector, which measures the amount (by
unit of volatility) by which the return of a risky asset is expected to
outperform the known return on a risk-free asset
21
31. Risk Premia and Bond Prices
Suppose there are no arbitrage opportunities, then there exists a risk-
neutral measure pricing measure Q
Then the value in time t of a zero coupon bond, with a monetary unit
as par value and maturity t + τ is given by
P(Xt , τ) = EQ
t exp(−
ˆ t+τ
t
rsds)
Let γ denote the risk premia vector, which measures the amount (by
unit of volatility) by which the return of a risky asset is expected to
outperform the known return on a risk-free asset
Assumption 3 (Risk neutral measure)
The risk-neutral measure Q defined by the following Radon-Nikdm
derivative
dQ
dP t
=
n
j=1
exp −
ˆ t
0
γj dWj,s −
1
2
ˆ t
0
γ2
j ds (5)
21
32. Using Girsanov’s theorem, the dynamics of the factors under Q are
dxj,t = γj σj dt + κj (xj−1,t − xj,t )dt + σj dWQ
j,t (6)
22
33. Using Girsanov’s theorem, the dynamics of the factors under Q are
dxj,t = γj σj dt + κj (xj−1,t − xj,t )dt + σj dWQ
j,t (6)
Therefore, the dynamics of the factors remain similar (affine) under the
risk neutral measure. In a matrix way, we have:
dXt = κ(θQ
− Xt )dt + Σ1/2
dWQ
t , (7)
Where θQ
portray the long-run mean of the state vector under the risk
neutral measure and is given by:
θQ
r = θr −
γ1σ1
κ1
, θQ
2 = −
γ2σ2
κ2
, . . . , θQ
n = −
γnσn
κn
(8)
22
34. Proposition 3 (Zero Coupon Bond Price)
The price at date t of a zero-coupon bond with maturity τ is given by
a affine model to Xt :
P(Xt , τ) = exp[−b(τ)T
Xt − c(τ)] (9)
The coeficients b(τ) and c(τ) satisfy the system of ordinary
differential equations:
b (τ) = en − κT
b(τ) (10)
c (τ) = b(τ)T
κθQ
−
1
2
b(τ)T
Σb(τ) (11)
With initial conditions b(τ) = 0 y c(τ) = 0, where en denotes a vector
with the value one in the n− th position, and zero elsewere.
23
35. Closed Form
Proposition 4 (Valuation under constant premia)
The price loadings bj (τ) satisfy
bj (τ) =
ˆ τ
0
aj (τ )dτ =
n
i=j
αi,j (1 − e−κi τ
) (12)
The long-run levels of the state vector under the measure Q are
θQ
r = θr − γ1σ1/κ1
θQ
2 = θr − γ1σ1 − γ2σ2/κ2
...
θQ
n = θr −
n
i=1
γi σi /κi
24
36. Proposition 4 (Valuation under constant premia)
The intercept c(τ) is given by
c(τ) = θr κ1
n
i=1
αi,1 τ −
1 − e−κi τ
κi
−
n
j=1
γj σj
n
i=j
αi,j τ −
1 − e−κi τ
κi
−
n
j=1
σ2
j
2
n
i=j
n
k=j
αi,j αk,j τ −
1 − e−κi τ
κi
−
1 − e−κk τ
κk
+
1 − e−(κi +κk )τ
κi + κk
25
37. Proposition 4 (Valuation under constant premia)
The intercept c(τ) is given by
c(τ) = θr κ1
n
i=1
αi,1 τ −
1 − e−κi τ
κi
−
n
j=1
γj σj
n
i=j
αi,j τ −
1 − e−κi τ
κi
−
n
j=1
σ2
j
2
n
i=j
n
k=j
αi,j αk,j τ −
1 − e−κi τ
κi
−
1 − e−κk τ
κk
+
1 − e−(κi +κk )τ
κi + κk
The long-run yield y∞ = limτ→+∞[b(τ)T
Xt + c(τ)]/τ satisfies
y∞ = θr −
n
j=1
σ2
j
κ2
j
γj
σj
κj +
1
2
, (13)
25
38. Introduction
A Cascade Model for the Interest Rate
Dimension-Invariant Term Structures
Base specification
Data and Estimation
Empirical Results
Conclusions
26
39. Base specification
Assumption 4 (Adjustment Speeds)
The sequence of adjustment speeds follows a geometric progression:
κj = kbj−1
, j = 1, 2, . . . , n. (14)
where b > 1 and k > 0 are constant real numbers.
27
40. Base specification
Assumption 4 (Adjustment Speeds)
The sequence of adjustment speeds follows a geometric progression:
κj = kbj−1
, j = 1, 2, . . . , n. (14)
where b > 1 and k > 0 are constant real numbers.
Assumption 6 (Risk Premia)
Factors risk premia are identical: γj = γ for all j
Under the risk-adjusted measure, the long-term levels of the first factor
are θQ
= θr − γσ/k.
27
41. Base specification
Assumption 5 (Volatilities)
Standard Model Factors volatilities are identical: σj = σ > 0 for all j.
28
42. Base specification
Assumption 5 (Volatilities)
Standard Model Factors volatilities are identical: σj = σ > 0 for all j.
Sigma Variant Model Factors volatilities follow a geometric
progression adjusted by a new parameter s
σj = σb(j−1)s
> 0, j = 1, . . . , n
• The standard model is specified by 5 parameters (k, b, σ, θ, γ)
with θQ
= θr − γσ/k.
• The SV model has 6, (k, b, σ, θ, γ, s), adding the s parameter in
volatility.
28
43. Introduction
A Cascade Model for the Interest Rate
Dimension-Invariant Term Structures
Data and Estimation
Empirical Results
Conclusions
29
44. Data
Sample
• Sovereign zero coupon curve of Colombia with maturities of 1, 2,
4, 6, 7, 8, 10 and 15 years, extracted from Bloomberg.
• Data are weekly with cut on Friday, with the average rate
between bid and ask. Linear Interpolation.
• From September 23, 2005 to December 28, 2018.
• 704 weekly observations of each of the 8 time series, for a total
of 5632 observations.
30
45. Data
Sample
• Sovereign zero coupon curve of Colombia with maturities of 1, 2,
4, 6, 7, 8, 10 and 15 years, extracted from Bloomberg.
• Data are weekly with cut on Friday, with the average rate
between bid and ask. Linear Interpolation.
• From September 23, 2005 to December 28, 2018.
• 704 weekly observations of each of the 8 time series, for a total
of 5632 observations.
Robustness, two subsamples
1. Estimation: from September 23, 2005 to December 30, 2016, for
a total of 589 weekly observations for each time series.
2. Forecast: from January 6, 2017 to December 28, 2018, with a
total of 115 observations.
30
46. Table 1: Yield Rate Statistics Summary
Autocorrelation
Maturity Mean Desv. Std Skew. Kurt. 1
1 ao 5.695 1.849 1.007 -0.037 0.9937
2 aos 6.201 1.799 1.126 0.239 0.9958
4 aos 6.859 1.770 0.909 0.137 0.9904
6 aos 7.331 1.721 0.823 0.112 0.9964
7 aos 7.607 1.525 0.907 0.418 0.9975
8 aos 7.567 1.665 0.448 -0.752 0.9982
10 aos 7.776 1.599 0.807 0.112 0.9959
15 aos 7.953 1.429 0.681 0.029 0.9938
Average 7.077 1.669 0.8385 0.032 0.9952
Table 1 reports a statistical summary of weekly observations (with Friday mid-
quotes) of yield rates with maturities of 1,2,4,6,7,8,10 and 15 years. Each se-
ries contains 704 weekly observations from September 23, 2015 to December
28, 2018. Entries report: mean (Mean), standard deviation (Dev. Std), skewness
(Skew), excess kurtosis (Kurt) and first order weekly autocorrelation.
31
47. Space-State Model
• It’s assumed that bond prices are observed in time intervals ∆t.
• The yield rates relate the prices of the zero coupon bonds
through a logarithmic transformation of the function (9), obtaining
a measurement equation:
y(t, τ) =
b(τ)T
Xt + c(τ)
τ
+ et (15)
32
48. Space-State Model
• It’s assumed that bond prices are observed in time intervals ∆t.
• The yield rates relate the prices of the zero coupon bonds
through a logarithmic transformation of the function (9), obtaining
a measurement equation:
y(t, τ) =
b(τ)T
Xt + c(τ)
τ
+ et (15)
• The state propagation equation (or transition equation) is:
Xt = A + ΦXt−∆t + ΣX εt , (16)
Where ∆t = 1/52, Φ = exp(−κ∆t), In denotes an identity matrix
n -dimensional, A = (In − Φ) θ, {εt } is i.i.d N(0, In),
With:
• ΣX = σ2
∆tIn for the standard model.
• ΣX = σ2
j ∆tIn con σ2
j = σ2
b(j−1)s
for the SV model.
32
49. Estimation and Optimization
• Being a model with unknown factors and parameters, and
possessing the desirable conditions of linearity and Gaussian
behavior is estimated through a Kalman filter.
• Initial values are assumed ˆx1|0 y P1|0 where ˆx1|0 = θr , the long
term rate y P1|0 = σ2
2k I.
33
50. Estimation and Optimization
• Being a model with unknown factors and parameters, and
possessing the desirable conditions of linearity and Gaussian
behavior is estimated through a Kalman filter.
• Initial values are assumed ˆx1|0 y P1|0 where ˆx1|0 = θr , the long
term rate y P1|0 = σ2
2k I.
• Algorithm of Broyden Fletcher Goldfarb Shanno with limited
memory and boundaries or L-BFGS-B, optimizes the likelihood
function:
log L = −
NT
2
log(2π) −
1
2
T
t=1
det |Σe| −
1
2
T
t=1
et [Σe]
−1
et (17)
Where Σe is the covariance matrix of the measurement error and
et the estimation error of the measurement equation.
33
51. Figure 1: Estimation and Optimization Algorithm
Initial values
κ, 𝑏, 𝜎, 𝜃𝑟, γ, 𝜎𝑒
Maximum Likelihood for T obs.
Minimize the log likelihood func.
negative sum conditional to the values
generate by the filter.
Kalman Filter for n factors on T
observations.
Estimation: ො𝑥 𝑡, Σ 𝑒, 𝑒𝑡
Optimize L-BFGS-B. The function
value reaches its minimum?
Finish optimization. Values that
minimize the log likelihood negative sum
have been found.
Select new values
κ, 𝑏, 𝜎, 𝜃𝑟, γ, 𝜎𝑒
No
Yes
34
52. Introduction
A Cascade Model for the Interest Rate
Dimension-Invariant Term Structures
Data and Estimation
Empirical Results
Conclusions
35
53. In sample Estimation
Table 2: Estimated Parameters and Standard Errors bp
Standard Model
Parameters
n k b σ θ γ σ2+
e θQr
L
1 1.0572*** 0.0110 0.0702* -0.4315*** 0.0311*** 0.0702 -12148.49
(0.0233) (—-) (0.1076) (0.0412) (0.0023) (0.0044)
2 0.4977*** 1.4088*** 0.0312*** 0.0753*** -0.4341*** 0.03741*** 0.0761 -12120.65
(0.0029) (0.0028) (0.0053) (0.0110) (0.00001) (0.0050)
3 0.5795*** 1.4043*** 0.0106*** 0.0455*** -0.4317*** 0.0302*** 0.0456 -12755.38
(0.0011) (0.0010) (0.0007) (0.0013) (0.000005) (0.0026)
4 1.7900*** 1.4053*** 0.0305 0.0610* -0.4351*** 0.0612 0.3164 -13107.42
(0.1007) (0.0485) (0.0548) (0.0336) (0.0129) (0.7079)
5 1.3255*** 2.0069*** 0.02314 0.0644 -0.4345*** 0.0249*** 0.0646 -12796.83
(0.0909) (0.1139) (0.0343) (0.0721) (0.0005) (0.3373)
6 0.1975*** 1.1218*** 0.0187*** 0.0466 0.1290*** 0.0739 0.0464 -9287.79
(0.0522) (0.0442) (0.0476) (0.6501) (0.0004) (6.4866)
7 0.3771*** 1.8281** 0.0151 0.0572 -0.4105*** 0.0099 0.0574 -14163.49
(0.0798) (0.7829) (0.0211) (0.0973) (0.0025) (0.483)
8 0.4547*** 1.6104*** 0.0126*** 0.0379*** -0.4309*** 0.0370*** 0.0381 -11532.20
(0.0031) (0.0035) (0.0025) (0.0034) (0.00002) (0.0098)
Table 2 reports the maximum likelihood estimates and standard errors (in parentheses) of the standard model parameters presented
in Calvet (2018). Each row represents a set of parameters estimated with the cascade with n = 1 to n = 8 components. The L
column reports the maximum weekly average likelihood log value for each model. Asymptotically, the statistician has a standard
normal distribution. Significance level: * 10 %, ** 5 % and *** 1 %.
+ Both measurement errors and their deviations are found in bp.
36
54. Table 3: Estimated Parameters and Standard Errors bp
SV Model
Parameters
n k b σ θ γ σ2+
e s θQ
r L
1 0.2596*** 0.0257 0.0601*** -0.1406*** 0.7948*** 0.0605 -6862.94
(0.0256) (—-) (0.0253) (0.0118) (0.0012) (0.0030) (—-)
2 0.1234*** 1.1523*** 0.0122*** 0.0286 0.4575*** 0.3780*** 0.1890*** 0.0280 -9876.86
(0.0007) (.0001) (0.0019) (0.1924) (0.00002) (0.0025) (0.00001)
3 1.0768*** 2.3100*** 0.0261*** 0.0317*** -0.4314*** 0.1110*** 0.4989*** 0.0314 -10508.42
(0.0069) (0.0040) (0.0066) (0.0055) (0.00007) (0.0023) (0.0041)
4 0.3271*** 1.4317*** 0.0024 0.0805*** -0.4463*** 0.0130*** 0.5027*** 0.0805 -15171.54
(0.0202) (0.1635) (0.1041) (0.0155) (0.0577) (0.0075) (0.0519)
5 0.3373*** 1.4027*** 0.0108*** 0.0526* -0.4346*** 0.0103 0.5007*** 0.0528 -13998.85
(0.0466) (0.0601) (0.0057) (0.0307) (0.0494) (0.0750) (0.0770)
6 1.4325*** 1.3914*** 0.0441 0.0302 -0.4326*** 0.0099 0.5094*** 0.0309 -14192.72
(0.0827) (0.0571) (0.0271) (0.0245) (0.0787) (0.7834) (0.1013)
7 0.1422*** 1.1927*** 0.0014 0.0963*** 0.2779*** 0.0046 0.4909*** 0.0963 -15587.02
(0.0016) (0.0009) (0.0038) (0.0001) (0.00001) (0.2547) (0.0002)
8 0.2492*** 1.2806*** 0.0291* 0.0647** -0.1018*** 0.0094 0.3817*** 0.0650 -13291.66
(0.0060) (0.0267) (0.0152) (0.0274) (0.0306) (1.5207) (0.0215)
Table 3 reports the maximum likelihood estimates and standard errors (in parentheses) of the variant model sigmaj parameters. Each row
represents a set of parameters estimated with the cascade with n = 1 to n = 8 components, where the measurement error is in basic points.
The cL column reports the maximum weekly average likelihood log value for each model. Asymptotically, the statistician has a standard normal
distribution. Significance level: * 10 %, ** 5 % and *** 1 %.
+ Both measurement errors and their deviations are found at bp.
37
55. In sample Fit
• The Root Mean Square Error (RMSE)
• Determine how far the values generated by the real model move
away.
• The RMSE has the following formula:
RMSE = E yt,τ − ˆyt,τ yt,τ − ˆyt,τ (18)
38
56. Table 4: RMSE in bp Standard Model
n
Maturity 1 2 3 4 5 6 7 8
1 ao 0.5037 0.8222 0.6478 0.6681 0.4894 2.7089 0.9226 1.3220
2 aos 0.4569 0.6496 0.3949 0.2561 0.1153 2.5289 0.7779 1.0354
4 aos 0.4260 0.2942 0.0763 0.1182 0.1850 2.2392 0.5338 0.5584
6 aos 0.4245 0.0864 0.1884 0.2196 0.2692 1.9981 0.3657 0.2788
7 aos 0.3843 0.0955 0.2585 0.2126 0.2545 1.8555 0.3460 0.2165
8 aos 0.4140 0.1902 0.3140 0.2680 0.3036 1.7671 0.3029 0.2039
10 aos 0.4125 0.2751 0.3747 0.2756 0.3093 1.5476 0.1834 0.14283
15 aos 0.3907 0.4614 0.4546 0.2840 0.3091 1.0422 0.1572 0.2479
Average 0.4266 0.3593 0.3384 0.2878 0.2794 1.9609 0.4487 0.5014
Table 5: RMSE in bp SV Model
n
Maturity 1 2 3 4 5 6 7 8
1 ao 0.2778 0.7127 0.1975 0.3307 0.6080 0.6390 0.7046 0.4186
2 aos 0.2212 0.4690 0.1971 0.2139 0.5500 0.5233 0.5642 0.3800
4 aos 0.2411 0.1197 0.4391 0.0661 0.4355 0.2320 0.3726 0.3575
6 aos 0.2971 0.2862 0.5649 0.1109 0.3486 0.1797 0.2875 0.3765
7 aos 0.3047 0.4390 0.6081 0.1057 0.3566 0.1483 0.2548 0.4222
8 aos 0.3363 0.5540 0.6375 0.1662 0.3007 0.2135 0.2408 0.4062
10 aos 0.3549 0.7759 0.6786 0.1678 0.2154 0.2002 0.1837 0.3558
15 aos 0.3717 1.2278 0.7155 0.2016 0.1918 0.2241 0.1502 0.2936
Average 0.3006 0.5730 0.5048 0.1704 0.3758 0.2950 0.3448 0.3762 39
57. Introduction
A Cascade Model for the Interest Rate
Dimension-Invariant Term Structures
Data and Estimation
Empirical Results
Conclusions
40
58. Conclusions
• Estimated values for different dimensions fluctuate along the size
of the cascade, for both models. Possible causes: Linearization
of measurement equation, data interpolation, optimization.
41
59. Conclusions
• Estimated values for different dimensions fluctuate along the size
of the cascade, for both models. Possible causes: Linearization
of measurement equation, data interpolation, optimization.
• The standard model is more stable than the SV model, but the
latter is more consistent in results in the sample and out-of-the
sample (work in progress).
41
60. Conclusions
• Estimated values for different dimensions fluctuate along the size
of the cascade, for both models. Possible causes: Linearization
of measurement equation, data interpolation, optimization.
• The standard model is more stable than the SV model, but the
latter is more consistent in results in the sample and out-of-the
sample (work in progress).
• For this type of affine model, it is advisable to use at least 3 factors
for the adjustment in the term structure data in Colombia
41
61. Conclusions
• Estimated values for different dimensions fluctuate along the size
of the cascade, for both models. Possible causes: Linearization
of measurement equation, data interpolation, optimization.
• The standard model is more stable than the SV model, but the
latter is more consistent in results in the sample and out-of-the
sample (work in progress).
• For this type of affine model, it is advisable to use at least 3 factors
for the adjustment in the term structure data in Colombia
• Dimension-invariant affine models do not seem suitable for fore-
casting the dynamics of Colombian TS (work in progress).
However, can be readily applied to pricing, bond portfolio risk
management and risk premium analysis.
41