Forecasting EUR–BNR currency exchange

Part 1: Starting with R – installation : http://msprogrammer.serviciipeweb.ro/2016/10/03/starting-with-r-as-in-sqlserver-r-server-and-vs-package/

Part 2: Starting with R – language and data http://msprogrammer.serviciipeweb.ro/2016/10/10/starting-with-r-language-and-data/

Part 3: Starting with R – basic and statistics http://msprogrammer.serviciipeweb.ro/2016/10/17/starting-with-r-basic-and-statistics/

Part 4: Making diagrams with R ( and transformation of data) : http://msprogrammer.serviciipeweb.ro/2016/10/31/making-diagrams-with-r/

Part 5:  Forecasting currency exchange with R   http://msprogrammer.serviciipeweb.ro/2016/11/06/forecasting-eurbnr-currency-exchange/

Let’s suppose that we have data from previous post

The form of the data was:

head(EUR,10)

      EUR                Data   RealDate Year
1  4.4629 2016-03-09 00:00:00 2016-03-09 2016
2  4.4624 2016-03-11 00:00:00 2016-03-11 2016
3  4.4705 2016-03-15 00:00:00 2016-03-15 2016
4  4.4775 2016-03-16 00:00:00 2016-03-16 2016
5  4.4765 2016-03-17 00:00:00 2016-03-17 2016
6  4.4694 2016-03-22 00:00:00 2016-03-22 2016
7  4.4641 2016-03-23 00:00:00 2016-03-23 2016
8  4.4655 2016-03-24 00:00:00 2016-03-24 2016
9  4.4639 2016-03-25 00:00:00 2016-03-25 2016
10 4.4619 2016-03-28 00:00:00 2016-03-28 2016

After reading materials

http://www.statoek.wiso.uni-goettingen.de/veranstaltungen/zeitreihen/sommer03/ts_r_intro.pdf

http://a-little-book-of-r-for-time-series.readthedocs.io/en/latest/src/timeseries.html

https://www.otexts.org/fpp/8/7

http://stats.stackexchange.com/questions/14742/auto-arima-with-daily-data-how-to-capture-seasonality-periodicity

https://www.r-bloggers.com/time-series-analysis-using-r-forecast-package/

http://www.statmethods.net/advstats/timeseries.html

and understanding that time series is not for beginners, I decide to find R packages that can do the job for me “automatically” . I mean, I give the pervious data –give me the forecast

So this link proved to be useful: http://blog.revolutionanalytics.com/2013/06/learning-time-series-with-r.html 

So this is the code  to importing forecast package( to forecast) and xts – extended time series – to can specify dates .

install.packages("forecast")
install.packages("xts")
library("forecast")
library("xts")

Then the real code:

myts <- xts(eur$EUR, eur$RealDate)

To see the data

head(myts,3)

and the result:

             [,1]
1999-01-04 1.3062
1999-01-05 1.3169
1999-01-06 1.3155

Data from backwards

tail(myts,3)

and the result

            [,1]
2016-09-23 4.4470
2016-09-26 4.4504
2016-09-27 4.4495

Now to the forecasting: tried to generate a model with auto.arima
fit<- auto.arima(myts)

auto.arima result was a warning also

Warning message:
In auto.arima(myts) :
  Unable to fit final model using maximum likelihood. AIC value approximated

So I tried

fit <- ets(myts)

that worked without warnings.

So let’s try a forecast for 5 days:

forecast(fit,5)

and the result

   Point Forecast    Lo 80    Hi 80    Lo 95    Hi 95
3835       4.449745 4.428025 4.471464 4.416527 4.482962
3836       4.449989 4.419210 4.480769 4.402916 4.497062
3837       4.450234 4.412485 4.487983 4.392502 4.507966
3838       4.450478 4.406836 4.494120 4.383734 4.517223
3839       4.450723 4.401874 4.499572 4.376014 4.525431

( below the arima result

     Point Forecast    Lo 80    Hi 80    Lo 95    Hi 95
3835       4.449287 4.427748 4.470825 4.416347 4.482226
3836       4.448930 4.416600 4.481259 4.399485 4.498374
3837       4.448847 4.409238 4.488455 4.388270 4.509423
3838       4.448763 4.403960 4.493565 4.380244 4.517282
3839       4.448666 4.399278 4.498054 4.373133 4.524198

)

And this is the situation

RealValues (RV)	ETS	Dif ETS (ETS-RV)/RV*100	arima	Dif Arima (A-RV)/RV*100
4.4493	4.449745	0.0100%	4.449287	0.0003%
4.4514	4.449989	0.0317%	4.448930	0.0555%
4.4523	4.450234	0.0464%	4.448847	0.0776%
4.4484	4.450478	0.0467%	4.448763	0.0082%
4.4576	4.450723	0.1543%	4.448666	0.2004%

Apparently , ETS is pretty ok ( with the exception of first result)