The ti (Time Index) and tis (Time Indexed Series) classes in the 
package provide date arithmetic facilities and an alternative to the somewhat
inflexible ts class in the standard R stats package. 

Time Indexes (ti class)

A time index has two parts: a tif (Time Index Frequency) code and a period.  tif
codes all lie in the open interval (1000..5000) and the period is a
nonnegative number less than 1e10.  The ti encodes both, as for any ti z

unclass(z) == (tif(z) * 1e10) + period(z)

Each tif has a particular base period (the time period where period(z) == 0).
For example, the base period for an anndecember (annual December) ti is the
year ending December 31, 1599.  Periods before the base period cannot be
represented by instances of the ti class.

If x and y are ti objects with the same tif, then 

x - y == period(x) - period(y)

and

x + (period(y) - period(x)) == y 

are both TRUE, so you can use ti's for various calendrical calculations. 

A jul class is also provided for Julian date-time objects. The jul()
constructor can create a jul from an ssDate (spreadsheet date), a
POSIXct or POSIXlt object, a ti object, a decimal time (like 2007.5), a
yyyymmdd number, a Date, or anything that can be coerced to a Date by
as.Date.  The ymd() function and its derivatives (year(), month(), day(), etc.)
work on anything that jul() can handle.


Time Indexed Series (tis class)

The tis class maps very closely to the FAME (http://www.sungard.com/Fame/) 
database notion of what a time series is.  A tis (Time Indexed Series) is
vector or matrix indexed by a ti.  If x is a tis, then start(x) gives the ti
for the first observation, and [start(x) + k] is the ti for the k'th
observation, while end(x) gives the ti for the last observation.  

You can replace, say, the 5'th observation in a tis x by

xstart <- start(x)
x[xstart + 4] <- 42

and of course the [ operator also works with a ti.  So if you want the value
of the daily series x from July 3, 1998, you can get it with

x[ti(19980703, "daily")]

provided, of course, that ymd(start(x)) <= 19980703 <= ymd(end(x)).

Numerous methods for tis objects are provided:
> methods(class = "tis")
 [1] aggregate.tis*     as.data.frame.tis* as.matrix.tis*     as.tis.tis*       
 [5] as.ts.tis*         cbind.tis*         cummax.tis*        cummin.tis*       
 [9] cumprod.tis*       cumsum.tis*        cycle.tis*         deltat.tis*       
[13] diff.tis*          edit.tis*          end.tis*           frequency.tis*    
[17] lag.tis*           lines.tis*         Ops.tis*           points.tis*       
[21] print.tis*         RowMeans.tis*      RowSums.tis*       start.tis*        
[25] tifName.tis*       tif.tis*           time.tis*          [<-.tis*          
[29] [.tis*             ti.tis*            t.tis              window.tis*       

as well as tisMerge() and tisPlot().  The convert() function change create
series of different frequencies from a given series in ways both more powerful
and more flexible than aggregate() can.



Setting Default Frequencies: 

Like the FAME DBMS, the ti class has seven weekly frequencies for weeks ending on
Sunday, Monday, and it has 12 annual frequencies, for years ending on the last
day of each of the 12 months. There are also multiple biweekly, bimonthly and
semiannual frequencies.  
At any time you can use the function setDefaultFrequencies() to change which
actual frequencies the strings "weekly", "biweekly", "bimonthly",
"quarterly", "semiannual" and "annual" refer to.  Note (as shown by
args(setDefaultFrequencies)) that if you don't specify some other frequencies
to setDefaultFrequencies(), you'll get default weeks ending on Monday and
default biweeks ending on the second Wednesday.  I chose these because they
are the weeks and biweeks used most often (for money and reserves data) in my
section at the Federal Reserve Board.  You might want to use something like

setHook(packageEvent("tis", "onLoad"), tis::setDefaultFrequencies(yourArgsGoHere))

in your .First() to automatically set the defaults up the way you want them
whenever this package is loaded.


LAGS:

The stats::lag(x, k) function says that a series lagged by a positive k starts
earlier.  The opposite is true for the Lag function in this package, to
maintain consistency with the common usage of 'first lag, second lag' and so
on in econometrics.

tisFilter:

The stats::filter() function coerces it's argument to a "ts" time series and
returns a "ts".  For "tis" and "zoo" series, this is not right.  Both I and
the author of the zoo package have requested that stats::filter be made an S3
generic with the current version renamed as filter.default.  This would allow
zoo and tis to define filter.zoo and filter.tis such that filter(aZooOrTis)
would do the right thing.  We are hopeful that this will happen soon.
Meanwhile, tisFilter() should be used to filter() a tis.