Time Series Questions
Hi there,
I am a graduate student with statistics background and recently got a quant internship interview. It is about Time Series. Anyone can help me about this questions?
-
How do you detect that a time series is following an ARMA process ?
-
Cite 5 type of automated strategy and give a brief description.
-
Give a description of a possible strategy to be long volatility in trading only the stock.
-
What kind of asset is usually modelled by a GARCH process ? Give a brief description.
I think I can answer three of the four.
You detect an ARMA process by examining the autocorrelation and partial autocorrelation function. If the ACF and PACF do not cut off after a specific lag, then you have an ARMA process. Of course, you need to make sure that you have a stationary process first before.
One way I see of long volatility is buying an option on a stock. As volatility increases, the value of the option increases. If you wanted to short volatility, you would write an option.
I just googled GARCH process -- http://www.studyfinance.com/jfsd/pdffiles/v13n2/johnston.pdf
GARCH processes are used to model time series data with varying variances. So basically any financial asset will exhibit varying volatility.
Absolutely agree with the long volatility answer. Long straddle.
Thank you TPride and ExplorerMan~~
Time Series Analysis (Originally Posted: 04/14/2012)
Interested in learning how time series analysis / statistics can be used to make trading models (detect/predict trading patterns, trends, etc). Anyone have any books they would suggest? Nothing that's too technical and complex as I'm not a quant, but enough substance so it could be helpful.
Econometrics books are pretty useful. Any college-level statistics books would also work.
Time series are pretty hard, to be honest. I'm just finishing up a course on the basics, a decent econometric grounding (like at the level of a college class) would be pretty important. Ok, I might be exaggerating a little bit, but since you just need to know is just DO OLS (and things like polynomial regression, probit, logit really aren't that necessary). I think that would be a good place to start. I can't recommend a book though since my professor hated everything and wrote his own coursepack that he distributed for the students.
Ruey Tsay wrote an easy enough book that should be comprehensible to most undergrads with a decent amount of math under their belt, but this is sort of like "I want to understand fundamental analysis and get good trade ideas despite not having an in-depth knowledge of accounting." You'll have to get down and dirty with the math at some point if you want to actually understand where your models work and don't.
So much this.
I am taking a statistical signal processing course right now, and to understand it you will need some serious math background. Not sure how much value a 35,000 ft summary will help you.
You have a estimation of the parameters of a model, how much risk are you willing to accept that the estimation is wrong, if it is within your threshold you keep it and update your model, if it isn't within the threshold you toss it into the bit bucket and move on.
Given your inaccurate description of a time-series course, I think you'd find machine learning far more beneficial.
Here ya go: http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/…
[quote=protectedclass]Here ya go: http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/…]
That might be too hard. Maybe start with:
http://ocw.mit.edu/courses/economics/14-384-time-series-analysis-fall-2…
Mathematical Economics and Time Series Help (Originally Posted: 12/05/2013)
Due to a change in responsibilities have to get up to speed on times series analysis (model building with macro data to forecast financial variables). Some econ/stats background but haven't taken a math course in year. I'm thinking the best route is math for econ review, then a time series book, then practicing with R itself but i'm really out of my element. Hoping to become proficient in a year (unless wildly unrealistic). Any suggestions on course of study or material would be greatly appreciated.
Do you know any econometrics? If not, I'd get yourself familiar with OLS as step 1.
Sorry should have said ive taken a stats, basic econometrics, and macro course in last couple years but no math courses for a long time
Do this in order:
If your goal is to build macro forecasting model, I fail to see how mathematical econ would be highly beneficial, but this book will get you up to speed on the general math: http://www.Amazon.com/Fundamental-Methods-Mathematical-Economics-Wainwright/dp/0070109109/ref=sr_sp-atf_title_1_2?s=books&ie=UTF8&qid=1386268800&sr=1-2&keywords=mathematical+economics
You're doing the right thing by learning R and I think it would be ideal to learn time-series + R together rather than sequentially if you wish to focus on applied macro research. If you wanted to achieve this in the minimal amount of time:
1) R Cookbook by Paul Teetor: http://www.Amazon.com/Cookbook-OReilly-Cookbooks-Paul-Teetor/dp/0596809158/ref=tmm_pap_title_0?ie=UTF8&qid=1386268939&sr=1-1 Go through: Ch 2,4,5,9,10,11 (imp.) and 14 ( very imp.)
2) Analysis of Financial Time Series by Ruey Tsay: http://www.Amazon.com/Analysis-Financial-Wiley-Probability-Statistics/dp/0470414359/ref=sr_sp-atf_title_1_1?s=books&ie=UTF8&qid=1386269125&sr=1-1&keywords=financial+time+series Go through: Ch 1-4,8,9
Great minds think alike, apparently.
haha yeah, though I do believe OP can jump into the deep end within 3 weeks and build a decent forecasting model given the right data sets, and then supplement some additional theory over time in order to build incrementally better models.
Silver bananna's to you both for the very useful advice. I guess the reason for the mathematical finance is BC I have a poor math background but maybe it's superflous to understand the math when you just need to apply the statistical software? Also I think the econometrics review would be help as I didn't do too well in those courses as opposed to diving right in to R. However, you're much more well versed in applied macro research than I am. Thoughts?
Obviously don't worry about the nitty-gritty - you don't need to know the EXACT definition of a beta estimator or the linear algebra reasoning behind multicollinearity. Just get a grasp of the basics and then go on to practicing with R.
That's exactly it. Building a macro forecasting model, particularly if you're attempting to forecast currencies, rates and commodities using fundamental+macro data, is theoretically supposed to be impossible, so at the very least you should assume that it ought to be difficult, because it absolutely is. However, the difficulty certainly isn't in the statistical aspects as even the most complex models can be run in 10 lines of code or less, and there are very few modeling techniques for you to deploy so its simply a matter of parameterization to create finite permutations of various models. A more difficult task would be to learn R as it is known to have a steep learning curve, and in doing so, you will be forced to learn regression modelling and time-series from a practical perspective, though you may have to live with the self-doubt of not understanding the intricacies of those models, which nine times out of ten only serves to quench your curiosity and nothing more. For example, I recently built a model that uses generalized splines to accurately provide 3m forecasts of the commodity currencies basket, and I only have a vague idea of how generalized splines work but I did understand the scenarios under which they were useful. Once you gain ground on a few modeling techniques and diagnostic tests, your macro domain expertise and programming knowledge undoubtedly take precedence over your statistical knowledge. So I think you should start by learning R programming in general, then transition towards learning how to run & interpret results of statistical, regression and times-series models in R and finally fill any grey areas with the appropriate literature if you need to. Despite what your professors would want you to believe having a very strong econometric background doesn't imply that you'll be capable of building sound macro forecasting models.
I think it might be a waste of time for this to go over much math. Probably best to go through a basic econometrics textbook (another good one and it's very clear/easy is Guide to Econometrics by Peter Kennedy), go through the Time Series book by Tsay, go through an R book, and just dive right into whatever it is you're trying to do and experiment and reference relevant items along the way. Good to know the basic concepts beforehand, but I've found you learn most by doing not by reading.
MacroArb and Chic have provided good advice.
Given your time constraints, I would skip Hayashi and Mostly Harmless. The former is too rigorous (and theoretical) for someone without a strong math background. Mostly Harmless is mostly for microeconometrics. Alpha-Chiang is a great book that will cover 90% of the math you will need, starting from high school level algebra.
There is no reason why couldn't be able to start working with R today, and in fact the best way to learn is by doing. There a bunch of free online resources for it. Start doing some basic matrix algebra and OLS regressions, and work your way up to fancier techniques as you learn them.
Learning to do all of this in one year should be no problem if you're dedicated.
Financial Time Series (Originally Posted: 03/20/2015)
Hello, can someone tell what is a good place to start reading about financial time series? I am fairly conversant with Econometrics and would love to start learning financial time series and how it ties with automated trading, if at all.
Basic Econometrics, Damodar N. Gujarati
HELP Time-Series Analysis!! (Originally Posted: 01/11/2012)
Hi Everyone:
I'm looking to improve my time-series forecasting.
Are there any other time series models I can to forecast macroeconomic events asides from error correction model, vector autoregressive, state space, smooth moving averages, arch & garch.
I need models that would allow for intervention analysis but allow for dummys to coincide & be non-linear.
At this stage in the game can you please just list other time series models not covered above and I can figure it out from there? Thank you for your time.
As as tokin of appreciation I will award sb's for each time series model listed above. Thanks again.
well....
the big joke is that none of the can predict massive swings ex ante; they work slightly better ex-post, but all you listed were frequentist models....
But! I didn't read the characteristics you actually wanted, but here goes:
bayesian VAR, bayesian VAR with stochastic volatility, bayesian VAR with stochastic vol and time-varying parameter drift, markov regime switching models, threshold VAR....
if you want shit from other fields, artificial neural networks, endless signal processing literature, classical Fourier series, etc.
if you want something totally badass, try bayesian dynamic model selection (DMS) or bayesian dynamic averaging (DMA)....though what ends up happening is that this is computationally expensive and you really don't create that much intellectual value added because you switch underlying factors and models within a larger bayesian framework....
good luck!
Sweet. Thanks a lot. I'll have to do some research. SBs for you.
Bump
Just curious: why'd you post this in the Consulting forum?
Reprehenderit et omnis soluta vel. Nulla et ratione quia. Natus voluptas aut enim officiis magnam illum provident. Est qui fugit dolorem repellendus eligendi. Consectetur quo molestiae sit enim.
Non labore voluptates doloribus repudiandae quaerat ab. Est vero et culpa. Et corporis optio quod explicabo dolor ipsum. Sint voluptas deleniti beatae vel. Hic corrupti vitae sit.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...
Autem explicabo sit expedita corrupti consequatur non. Omnis facere soluta non. Consequatur consequatur qui et qui et iusto ea modi. Est laboriosam ex a velit vero mollitia iure. Sed aut modi ipsum aperiam et velit iste. Natus illo voluptate vel autem enim dolorum ducimus nulla.
Optio et sit saepe natus. Impedit neque asperiores quae eveniet architecto. Aperiam natus rerum saepe consequatur qui et.
Qui aspernatur aut et autem iste et. Et voluptatem reiciendis dolorum. Velit labore fugiat quia porro impedit excepturi incidunt dolorem. Magni provident velit culpa explicabo. Repellat dolorum cupiditate nulla enim. Nulla minima iusto cumque adipisci quidem consequatur. Molestias odit delectus veritatis laboriosam incidunt.
Eum est ut dolore. Odit repellendus est et nisi maxime excepturi. Eius harum voluptas esse labore voluptas dicta veniam et. Tenetur voluptate et qui id earum laborum. Porro occaecati nihil iusto doloribus.