hreg: Four Parameter Harmonic Regression
In CliffordLai/harper: Harmonic Regression and Periodicity Test

Description Usage Arguments Details Value Note Author(s) See Also Examples

Estimates mu, A, B and lambda in the harmonic regression, y(t)=mu+A*cos(2*pi*lambda*t)+B*sin(2*pi*lambda*t)+e(t), where e(t) is assumed IID mean zero and constant variance. Four estimation methods are available least-squares (LS), least absolute deviation (L1) and a robust regression methods (MM).

1 2	hreg(z, method = c("LS", "L1", "MM"), K = 500, ncpu = 1, exactQ = FALSE)

`z`	series.
`method`	one of "LS", "L1", "MM"/
`K`	number of subintervals.
`ncpu`	default 1.
`exactQ,`	default setting is FALSE. It indicates whether nolinear optimizer is used to obtain the final estimate of the frequency.

The recommended setting of ncpu is 1 - see note. The default setting for K is the 500 or the number of Fourier frequencies - whichever is larger. LS regression uses the QR algorithm directly to efficiently compute the residual sum of squares. L1-regression is fitted using the function rq.fit.br in the package quantreg. Robust regression uses the robustbase package function lmrob. The parallel package may be used to improve the efficiency for long time series but when the series length is less than 100 using ncpu=1 with K=500 is much faster due to the overhead - despite using splitIndices to balance the load on each compute note.

list with the following components

coefficients const, A, B, lambda
residuals
periodogram
z
t
K
model
lambdaRange

The parallel package may be used to speed up the computation on the frequency grid but on Windows machines the overhead for computation is so large that this is not usually effective unless the series length is quite long. See nottem in the example section. It was surprising that the overhead was so large despite using splitIndices to balance the load on each cpu.

Yuanhao Lai and A. I. McLeod

qr

z<-c(0.42, 0.89, 1.44, 1.98, 2.21, 2.04, 0.82, 0.62, 0.56, 0.8, 1.33)
hreg(z)
#
#on multicore pcs, more the package parallel may be used for the
#grid computation but unless n is very large this is not recommended.
## Not run:  #adjust ncpu
system.time(ans1 <- hreg(z)) #0.06 sec on my computer
system.time(ans2 <- hreg(z, ncpu=8)) #1.67 sec 
system.time(ans3 <- hreg(z, method="L1")) #0.06 sec on my computer
system.time(ans4 <- hreg(z, method="L1", ncpu=8)) #4.11 sec
system.time(ans5 <- hreg(z, method="MM")) #2.89 sec on my computer
system.time(ans6 <- hreg(z, method="MM", ncpu=8)) #2.65 sec 

## End(Not run)