7,173 reputation
11142
bio website wiki.scn.sap.com/wiki/pages/…
location Switzerland
age 39
visits member for 4 years, 4 months
seen 13 hours ago

During the day, I create software to forecast retail sales and calculate order proposals. Lots of time series, with an emphasis on fast, automatic and robust data cleansing and forecasting. I'm an elected director for the International Institute of Forecasters and an Associate Editor for their practitioner-oriented forecasting journal Foresight. At night, I switch hats and do inferential statistics for academic clinical and biological psychology. My main tool is the statistical computing environment R.

profile for Stephan Kolassa on Stack Exchange, a network of free, community-driven Q&A sites


Jan
28
comment Daily Ticket Sales
You're welcome. Feel free to upvote and/or accept my answer if it helped you.
Jan
28
answered Daily Ticket Sales
Jan
28
revised Daily Ticket Sales
added 7 characters in body
Jan
28
comment Missing F Statistic in Regression Output
Please edit your question to clarify, e.g., to include your regression call and output.
Jan
28
revised Forecasting in Stata
added time-series tag
Jan
27
revised Daily Ticket Sales
included sample data from comments
Jan
27
comment Best method for short time-series
You're welcome ;-) I guess the takeaway is that "short" is very context-dependent: for sensor reading series or in finance, 1000 data points is "short" - but in supply chain management, 20 monthly observations is almost normal, and "short" will only start at 12 or fewer observations.
Jan
27
comment Best method for short time-series
+1 to @Scortchi's comment. Incidentally, out of the 3,003 M3 series (available in the Mcomp package for R), 504 have 20 or fewer observations, specifically 55% of the yearly series. So you could look up the original publication and see what worked well for yearly data. Or even dig through the original forecasts submitted to the M3 competition, which are available in the Mcomp package (list M3Forecast).
Jan
27
comment Best method for short time-series
[2,3,4] do not mention short time series, and look at the plots in [2]: >120 observations. [4] concentrates on finance, where you have enormously more than 20 observations. [5] writes about "short time series, typically 1,000 points long" (p. 216). I see no way to reliably and robustly fit a TAR or similar model, or any of the more complex ones you link to, with <20 observations. (BTW: I also do some inferential statistics on the side, and with fewer than 20 observations, you really can't estimate more than the mean and one more parameter.)
Jan
27
answered Best method for short time-series
Jan
27
comment Best method for short time-series
+1 for the link to Rob Hyndman's post. (However, I am tempted to -1 for the complex models. I'd be extremely wary of using threshold or any other nonlinear time series methods on time series of less than 20 observations. You are almost certain to overfit, which goes directly counter to the OP's requirement of a robust method.)
Jan
27
revised Daily Ticket Sales
added forecasting tag and removed "thanks"
Jan
14
comment How to take advantage of multiples series with the same behaviour for forecasting?
He is. He co-authored both the open source textbook you linked to and the hierarchical time series approach, and he created the hts package for R for hierarchical time series - no python implementation, though - and he's editor-in-chief of the IJF and he posts here frequently. Take him seriously.
Jan
14
comment Male and Female Chess Players - Expected Discrepancies at Tails of Distributions
"As I'm contradicting the results of a published paper, I guess I must ask - what am I doing wrong?" - publication is no guarantee of correctness...
Jan
14
revised How to take advantage of multiples series with the same behaviour for forecasting?
added time-series tag
Jan
14
comment How to take advantage of multiples series with the same behaviour for forecasting?
+1. This approach has been found to improve forecast accuracy across the hierarchy, see here and here.
Jan
13
revised function for calculating the mean of a chosen row?
formatting
Jan
13
revised glm function in R uses conditional or unconditional likelihood?
added R tag
Jan
12
comment Fitting a quadratic through 5 points, goal is to find the maximum
In the stratified bootstrap, you don't sample one y over each x, but multiple ones (with replacement). With my toy data, I have 10 y over each x. So the stratified bootstrap would sample 10 times from the y vector over each x (with replacement). So you again have 50 data points to fit the model. The result is that the parameter estimates (and the estimate for the x coordinate of the maximum) will have a lower variance than if you only sample one y over each x. (You may want to read up on the bootstrap.)
Jan
12
revised Fitting a quadratic through 5 points, goal is to find the maximum
added 11 characters in body