SLIDE 1
FMSN60/MASM18 Financial Statistics Lecture 1, Introduction and stylized facts
Magnus Wiktorsson
FMSN60/MASM18 Financial Statistics Lecture 1, Introduction and - - PowerPoint PPT Presentation
FMSN60/MASM18 Financial Statistics Lecture 1, Introduction and - - PowerPoint PPT Presentation
FMSN60/MASM18 Financial Statistics Lecture 1, Introduction and stylized facts Magnus Wiktorsson People and homepage 222 86 25, MH:130 (Lecturer) samuel.wiqvist@matstat.lu.se , 222 79 83, MH:326 (Computer exercises) 222 04 85 , MH:223 (Computer
SLIDE 2
SLIDE 3
Purpose:
▶ The course should provide tools for analyzing data, and use these tools in combination with economic theory. The main applications are valuation and risk management. ▶ The course is intended to provide necessary statistical tools supporting courses like ’EXTQ35 Financial Valuation and Risk Management’ or ’FMSN25/MASM24 Valuation of Derivative Assets’.
SLIDE 4
Inference problems?
▶ Forecast prices, interest rates, volatilities (under the P and Q measures) ▶ Filtering of data (e.g. estimating hidden states such as stochastic volatility or credit default intensity) ▶ Distribution of prediction errors; can we improve the model? ▶ What about extreme events? How do we estimate parameters in general models?
Cross covariance and auto covariance. Often results in Non-linear, Non-Gaussian, Non-stationary models...
SLIDE 5
Inference problems?
▶ Forecast prices, interest rates, volatilities (under the P and Q measures) ▶ Filtering of data (e.g. estimating hidden states such as stochastic volatility or credit default intensity) ▶ Distribution of prediction errors; can we improve the model? ▶ What about extreme events? ▶ How do we estimate parameters in general models?
▶ Cross covariance and auto covariance. ▶ Often results in Non-linear, Non-Gaussian, Non-stationary models...
SLIDE 6
Example I – Daily interest data - big crisis in Sweden during the early 1990s
See Section 2.4 in the book for more information.
Q1−92 Q2−92 Q3−92 Q4−92 Q1−93 10 100 500 STIBOR and REPO Yields 1992 REPO STIBOR 1W STIBOR 1M STIBOR 3M STIBOR 6M
Forecasts - 0.5 % or 500 %? Covariation with of market factors? - Can this happen again? Models and distributions.
SLIDE 7
Example I – Daily interest data - big crisis in Sweden during the early 1990s
See Section 2.4 in the book for more information.
Q1−92 Q2−92 Q3−92 Q4−92 Q1−93 10 100 500 STIBOR and REPO Yields 1992 REPO STIBOR 1W STIBOR 1M STIBOR 3M STIBOR 6M
▶ Forecasts - 0.5 % or 500 %? ▶ Covariation with of market factors? - Can this happen again? ▶ Models and distributions.
SLIDE 8
Electricity spot price and Hydrological situation
SLIDE 9
Example II – Forward prices on Nordpool
▶ Traders are interested in predicting price movements on the futures on Nordpool on yearly contracts. ▶ Or predicting the movements on short horizons (days or weeks). ▶ Expected movement and/or prob. of declining prizes. ▶ What about fundamental factors?
- 1. Hydrological situation is the energy stored as
snow, ground water or in reservoirs
- 2. Time to maturity.
- 3. Perfect or imperfect markets.
▶ Other factors – suggestions?
SLIDE 10
Ex – Forwards on Nordpool, contd.
There is a strong dependence between the hydrological situation and the price. ▶ How do we model this dependence, e.g. what model should we use? Is the relation linear? ▶ How do we fit the chosen model? ▶ How do we know if the model is good enough? ▶ One supermodel or several models? ▶ Adaptive models?
SLIDE 11
Contents
The course treats estimation, identification and validation in non-linear dynamical stochastic models for financial applications based on data and prior knowledge. There are rarely any absolutely correct answers in this course, but there are often answers that are absolutely wrong. This was expressed by George Box as
All models are wrong - but some are useful!
Think for yourself, and question the course material!
SLIDE 12
Contents
The course treats estimation, identification and validation in non-linear dynamical stochastic models for financial applications based on data and prior knowledge. There are rarely any absolutely correct answers in this course, but there are often answers that are absolutely wrong. This was expressed by George Box as
All models are wrong - but some are useful!
Think for yourself, and question the course material!
SLIDE 13
Contents, 2
Discrete and continuous time. ▶ Parameter estimation (LS, ML, GMM, EF), model identification and model validation. ▶ Modelling of variance, ARCH, GARCH, ..., and
- ther approaches.
▶ Stochastic calculus and SDEs. ▶ State space models and filters Kalman filters (and versions thereof) and particle filters
SLIDE 14
Course goals -Knowledge and Understanding
For a passing grade the student must: ▶ handle variance models such as the GARCH family, stochastic volatility, and models use for high-frequency data, ▶ use basic tool from stochastic calculus: Ito’s formula, Girsanov transformation, martingales, Markov processes, filtering, ▶ use tools for filtering of latent processes, such as Kalman filters and particle filters, ▶ statistically validate models from some of the above model families.
SLIDE 15
Course goals -Skills and Abilities
For a passing grade the student must: ▶ be able to find suitable stochastic models for financial data, ▶ work with stochastic calculus for pricing of financial contracts and for transforming models, ▶ understand when and how filtering methods should be applied, ▶ validate a chosen model, ▶ solve all parts of a modelling problem using economic and statistical theory (from this course and from other courses) where the solution includes model specification, inference, and model choice, ▶ utilise scientific articles within the field and related fields. ▶ present the solution in a written technical report, as well as orally,
SLIDE 16
Literature
▶ Lindström, E., Madsen, H., Nielsen, J. N. (2015) Statistics for Finance, Chapman & Hall, CRC press. ▶ Handouts (typically articles on the course home page) Course program.
SLIDE 17
Properties of financial data
▶ No Autocorrelation in returns ▶ Unconditional heavy tails ▶ Gain/Loss asym. ▶ Aggregational Gaussianity ▶ Volatility clustering ▶ Conditional heavy tails ▶ Significant autocorrelation for abs. returns - long range dependence? ▶ Leverage effects ▶ Volume/Volatility correlation ▶ Asym. in time scales Evaluate claims on S&P 500 data.
SLIDE 18
Autocorrelation in returns
10 20 30 40 50 60 70 80 90 100 Lag
- 2
2 4 6 8 10 10 -5 Covariance log returns
No or little autocorrelation.
SLIDE 19
Unconditional distribution
- 0.2
- 0.15
- 0.1
- 0.05
0.05 0.1 Data 0.001 0.003 0.01 0.02 0.05 0.10 0.25 0.50 0.75 0.90 0.95 0.98 0.99 0.997 0.999 Probability Normplot unconditional log returns
Normplot of the unconditional returns.
SLIDE 20
Gain/Loss asym.
1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 500 1000 1500 2000 2500 3000 S & P 500
Losses are larger than gains (data is Index). This may contradict the EMH, see Nystrup et al. (2016).
SLIDE 21
- Aggr. Gaussianity
- 0.2
- 0.15
- 0.1
- 0.05
0.05 0.1 Data 0.001 0.003 0.01 0.02 0.05 0.10 0.25 0.50 0.75 0.90 0.95 0.98 0.99 0.997 0.999 Probability Normplot log returns Daily
- 0.3
- 0.25
- 0.2
- 0.15
- 0.1
- 0.05
0.05 0.1 0.15 Data 0.001 0.003 0.01 0.02 0.05 0.10 0.25 0.50 0.75 0.90 0.95 0.98 0.99 0.997 0.999 Probability Normplot log returns Monthly
- 0.3
- 0.2
- 0.1
0.1 0.2 0.3 Data 0.001 0.003 0.01 0.02 0.05 0.10 0.25 0.50 0.75 0.90 0.95 0.98 0.99 0.997 0.999 Probability Normplot log returns Quarterly
- 0.4
- 0.3
- 0.2
- 0.1
0.1 0.2 0.3 Data 0.01 0.02 0.05 0.10 0.25 0.50 0.75 0.90 0.95 0.98 0.99 Probability Normplot log returns Yearly
Returns are increasingly Gaussian. Interpretation?
SLIDE 22
- Vol. Clustering
1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
- 0.25
- 0.2
- 0.15
- 0.1
- 0.05
0.05 0.1 0.15 S & P 500 log returns
Volatility clusters. Average cluster size?
SLIDE 23
Dependence in absolute returns
10 20 30 40 50 60 70 80 90 100 Lag
- 1
1 2 3 4 5 6 10 -5 Covariance absolute log returns
Significant autocorrelation. Long range dependence
- r other reason? Hint: Nystrup et al., (2015, 2016)
SLIDE 24
Conditional distribution
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
- 12
- 10
- 8
- 6
- 4
- 2
2 4 6 8 Conditional log returns
- 10
- 8
- 6
- 4
- 2
2 4 6 Data 0.001 0.003 0.01 0.02 0.05 0.10 0.25 0.50 0.75 0.90 0.95 0.98 0.99 0.997 0.999 Probability Normplot conditional log returns
Normplot of the conditional returns (GARCH(1,1) filter).
SLIDE 25
No correlation in conditional absolute returns
10 20 30 40 50 60 70 80 90 100 Lag
- 0.05
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 Covariance cond absolute log returns
SLIDE 26
Leverage effects
▶ Most assets are negatively correlated with any measure of volatility. ▶ One popular explanation is corporate debt. ▶ Makes sense if you are risk averse.
SLIDE 27
Volume/Volatility correlation
▶ Trading volume is correlated with the volatility. ▶ Sometimes modelled with ’business time’ in
- ption valuation community - cf. Time Shifted
Levy processes models, Def 7.12, such as NIG-CIR model.
SLIDE 28
- Asym. in time scales
▶ Coarse-grained measurements can predict fine scaled volatility ▶ While fine scaled volatility have difficulties predicting coarse scale volatility
SLIDE 29
Extra material
Feel free to download the paper (you need a Lund University IP address or STIL login - to access the paper.)
▶ Cont, R. Empirical properties of asset returns: stylized facts and statistical issues. Quantitative Finance, Vol. 1, No. 2 (March 2001) 223-236. http://ludwig.lub.lu.se/login?url=http: //dx.doi.org/10.1080/713665670 ▶ Nystrup, P., Madsen, H., & Lindström, E. (2015). Stylised facts of financial time series and hidden Markov models in continuous time. Quantitative Finance, 15(9), 1531-1541. http://ludwig.lub.lu.se/login?url=http: //dx.doi.org/10.1080/14697688.2015.1004801 ▶ Nystrup, P., Madsen, H., & Lindström, E. (2016). Long Memory of Financial Time Series and Hidden Markov Models with Time-Varying Parameters. Journal of Forecasting, http://ludwig.lub.lu.se/login?url=http: //dx.doi.org/10.1002/for.2447
SLIDE 30