Time Series Analysis 'OpenBudgets.eu'

Estimate and return the needed parameters for visualizations designed for 'OpenBudgets.eu' < http://openbudgets.eu/> time series data. Calculate time series model and forecast parameters in budget time series data of municipalities across Europe, according to the 'OpenBudgets.eu' data model. There are functions for measuring deterministic and stochastic trend of the input time series data with 'ACF', 'PACF', 'Phillips Perron' test, 'Augmented Dickey Fuller (ADF)' test, 'Kwiatkowski-Phillips-Schmidt-Shin (KPSS)' test, 'Mann Kendall' test for monotonic trend and 'Cox and Stuart' trend test, decomposing with local regression models or 'stl' decomposition, fitting the appropriate 'arima' model and provide forecasts for the input 'OpenBudgets.eu' time series fiscal data. Also, can be used generally to extract visualization parameters convert them to 'JSON' format and use them as input in a different graphical interface.


TimeSeries.OBeu

Kleanthis Koupidis, Charalampos Bratsas

TimeSeries.OBeu

Εstimate and return the necessary parameters for time series visualizations, used in OpenBudgets.eu. It includes functions to test stationarity (with ACF, PACF, Phillips Perron test, Augmented Dickey Fuller (ADF) test, Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test, Mann Kendall Test For Monotonic Trend and Cox and Stuart trend test), decompose, model and forecast Budget time series data of municipalities across Europe, according to the OpenBudgets.eu data model.

This package can generally be used to extract visualization parameters convert them to JSON format and use them as input in a different graphical interface. Most functions can have general use out of the OpenBudgets.eu data model. You can see detailed information here.

install.packages(TimeSeries.OBeu) 
# or
# alternatively install the development version from github
devtools::install_github("okgreece/TimeSeries.OBeu")

Load library TimeSeries.OBeu

library(TimeSeries.OBeu)

Time Series analysis in a call

ts.analysis is used to estimate autocorrelation and partial autocorrelation of input time series data, autocorrelation and partial autocorrelation of the model residuals, trend, seasonal (if exists) and remainder components, model parameters such as arima order, arima coefficients etc. and the desired forecasts with their corresponding confidence intervals.

ts.analysis returns by default a json object, if tojson parameter is FALSE returns a list object and the default forecast step is set to 1.

results = ts.analysis(Athens_executed_ts, prediction.steps = 2, tojson=TRUE) # json string format
jsonlite::prettify(results) # use prettify of jsonlite library to add indentation to the returned JSON string
## {
##     "acf.param": {
##         "acf.parameters": {
##             "acf": [
##                 1,
##                 0.5302,
##                 0.2018,
##                 -0.1397,
##                 -0.4059,
##                 -0.3556,
##                 -0.3939,
##                 -0.073,
##                 0.071,
##                 0.0676,
##                 0.0285
##             ],
##             "acf.lag": [
##                 0,
##                 1,
##                 2,
##                 3,
##                 4,
##                 5,
##                 6,
##                 7,
##                 8,
##                 9,
##                 10
##             ],
##             "confidence.interval.up": [
##                 0.5658
##             ],
##             "confidence.interval.low": [
##                 -0.5658
##             ]
##         },
##         "pacf.parameters": {
##             "pacf": [
##                 0.5302,
##                 -0.1102,
##                 -0.2817,
##                 -0.2903,
##                 0.0427,
##                 -0.2781,
##                 0.2318,
##                 -0.1163,
##                 -0.1829,
##                 -0.209
##             ],
##             "pacf.lag": [
##                 1,
##                 2,
##                 3,
##                 4,
##                 5,
##                 6,
##                 7,
##                 8,
##                 9,
##                 10
##             ],
##             "confidence.interval.up": [
##                 0.5658
##             ],
##             "confidence.interval.low": [
##                 -0.5658
##             ]
##         },
##         "acf.residuals.parameters": {
##             "acf.residuals": [
##                 1,
##                 0.8646,
##                 0.7284,
##                 0.6039,
##                 0.4589,
##                 0.3295,
##                 0.154,
##                 -0.0016,
##                 -0.1241,
##                 -0.2595,
##                 -0.3802,
##                 -0.5098,
##                 -0.6276,
##                 -0.5885,
##                 -0.5207,
##                 -0.4629
##             ],
##             "acf.residuals.lag": [
##                 0,
##                 1,
##                 2,
##                 3,
##                 4,
##                 5,
##                 6,
##                 7,
##                 8,
##                 9,
##                 10,
##                 11,
##                 12,
##                 13,
##                 14,
##                 15
##             ],
##             "confidence.interval.up": [
##                 0.5658
##             ],
##             "confidence.interval.low": [
##                 -0.5658
##             ]
##         },
##         "pacf.residuals.parameters": {
##             "pacf.residuals": [
##                 0.8646,
##                 -0.0756,
##                 -0.0325,
##                 -0.1597,
##                 -0.0335,
##                 -0.2937,
##                 -0.0528,
##                 -0.046,
##                 -0.162,
##                 -0.1372,
##                 -0.2201,
##                 -0.2078,
##                 0.4336,
##                 0.1187,
##                 -0.0519
##             ],
##             "pacf.residuals.lag": [
##                 1,
##                 2,
##                 3,
##                 4,
##                 5,
##                 6,
##                 7,
##                 8,
##                 9,
##                 10,
##                 11,
##                 12,
##                 13,
##                 14,
##                 15
##             ],
##             "confidence.interval.up": [
##                 0.5658
##             ],
##             "confidence.interval.low": [
##                 -0.5658
##             ]
##         }
##     },
##     "decomposition": {
##         "stl.plot": {
##             "trend": [
##                 488397393.1091,
##                 472512470.2805,
##                 473063423.4839,
##                 487284165.8281,
##                 519914575.4486,
##                 549044538.1581,
##                 546747322.3717,
##                 517885722.186,
##                 482561749.2963,
##                 453474237.5689,
##                 423909077.9758,
##                 393617768.3187
##             ],
##             "conf.interval.up": [
##                 525849686.8842,
##                 495462596.0604,
##                 495888427.6281,
##                 512171768.3925,
##                 545880538.4876,
##                 575706534.5412,
##                 573409318.7548,
##                 543851685.225,
##                 507449351.8607,
##                 476299241.7132,
##                 446859203.7557,
##                 431070062.0938
##             ],
##             "conf.interval.low": [
##                 450945099.334,
##                 449562344.5005,
##                 450238419.3396,
##                 462396563.2637,
##                 493948612.4097,
##                 522382541.7751,
##                 520085325.9887,
##                 491919759.147,
##                 457674146.7319,
##                 430649233.4247,
##                 400958952.1958,
##                 356165474.5436
##             ],
##             "seasonal": {
## 
##             },
##             "remainder": [
##                 3494473.6909,
##                 -6782427.4905,
##                 -360030.3839,
##                 -20859217.1881,
##                 8715868.0414,
##                 20321961.4419,
##                 -24805255.8217,
##                 12476896.984,
##                 -25628827.4663,
##                 18714394.8611,
##                 -9197723.8358,
##                 1891498.5713
##             ],
##             "time": [
##                 2004,
##                 2005,
##                 2006,
##                 2007,
##                 2008,
##                 2009,
##                 2010,
##                 2011,
##                 2012,
##                 2013,
##                 2014,
##                 2015
##             ]
##         },
##         "stl.general": {
##             "degfr": [
##                 5.4179
##             ],
##             "degfr.fitted": [
##                 5.1011
##             ],
##             "stl.degree": [
##                 2
##             ]
##         },
##         "residuals_fitted": {
##             "residuals": [
##                 3494473.6909,
##                 -6782427.4905,
##                 -360030.3839,
##                 -20859217.1881,
##                 8715868.0414,
##                 20321961.4419,
##                 -24805255.8217,
##                 12476896.984,
##                 -25628827.4663,
##                 18714394.8611,
##                 -9197723.8358,
##                 1891498.5713
##             ],
##             "fitted": [
##                 488397393.1091,
##                 472512470.2805,
##                 473063423.4839,
##                 487284165.8281,
##                 519914575.4486,
##                 549044538.1581,
##                 546747322.3717,
##                 517885722.186,
##                 482561749.2963,
##                 453474237.5689,
##                 423909077.9758,
##                 393617768.3187
##             ],
##             "time": [
##                 2004,
##                 2005,
##                 2006,
##                 2007,
##                 2008,
##                 2009,
##                 2010,
##                 2011,
##                 2012,
##                 2013,
##                 2014,
##                 2015
##             ],
##             "line": [
##                 0
##             ]
##         },
##         "compare": {
##             "resid.variance": [
##                 258964785711184
##             ],
##             "used.obs": [
##                 2004,
##                 2015,
##                 2009.5,
##                 2006.75,
##                 2012.25
##             ],
##             "loglik": [
##                 -1.42430632141151e+015
##             ],
##             "aic": [
##                 2.84861264282304e+015
##             ],
##             "bic": [
##                 2.84861264282304e+015
##             ],
##             "gcv": [
##                 789007326652162
##             ]
##         }
##     },
##     "model.param": {
##         "model": {
##             "arima.order": [
##                 2,
##                 1,
##                 0,
##                 0,
##                 1,
##                 1,
##                 0
##             ],
##             "arima.coef": [
##                 -0.2,
##                 0.304,
##                 0.1684
##             ],
##             "arima.coef.se": [
##                 0.5484,
##                 0.3034,
##                 0.5345
##             ]
##         },
##         "residuals_fitted": {
##             "residuals": [
##                 491891.5916,
##                 -24734053.8013,
##                 4848198.2869,
##                 2291242.4698,
##                 58442566.8183,
##                 45241384.4941,
##                 -65806529.3585,
##                 -2362504.0059,
##                 -56932278.3288,
##                 7600701.3147,
##                 -33386168.6854,
##                 -29710365.2918
##             ],
##             "fitted": [
##                 491399975.2084,
##                 490464096.5913,
##                 467855194.8131,
##                 464133706.1702,
##                 470187876.6717,
##                 524125115.1059,
##                 587748595.9085,
##                 532725123.1759,
##                 513865200.1588,
##                 464587931.1153,
##                 448097522.8254,
##                 425219632.1818
##             ],
##             "time": [
##                 2004,
##                 2005,
##                 2006,
##                 2007,
##                 2008,
##                 2009,
##                 2010,
##                 2011,
##                 2012,
##                 2013,
##                 2014,
##                 2015
##             ],
##             "line": [
##                 0
##             ]
##         },
##         "compare": {
##             "resid.variance": [
##                 1.96694555669531e+015
##             ],
##             "variance.coef": [
##                 [
##                     0.3007,
##                     0.0586,
##                     -0.2532
##                 ],
##                 [
##                     0.0586,
##                     0.0921,
##                     -0.029
##                 ],
##                 [
##                     -0.2532,
##                     -0.029,
##                     0.2857
##                 ]
##             ],
##             "not.used.obs": [
##                 0
##             ],
##             "used.obs": [
##                 11
##             ],
##             "loglik": [
##                 -207.6519
##             ],
##             "aic": [
##                 423.3037
##             ],
##             "bic": [
##                 424.8953
##             ],
##             "aicc": [
##                 429.9704
##             ]
##         }
##     },
##     "forecasts": {
##         "ts.model": [
##             "ARIMA(2,1,1)"
##         ],
##         "data_year": [
##             2004,
##             2005,
##             2006,
##             2007,
##             2008,
##             2009,
##             2010,
##             2011,
##             2012,
##             2013,
##             2014,
##             2015
##         ],
##         "data": [
##             491891866.8,
##             465730042.79,
##             472703393.1,
##             466424948.64,
##             528630443.49,
##             569366499.6,
##             521942066.55,
##             530362619.17,
##             456932921.83,
##             472188632.43,
##             414711354.14,
##             395509266.89
##         ],
##         "predict_time": [
##             2016,
##             2017
##         ],
##         "predict_values": [
##             376873927.6929,
##             374763602.0226
##         ],
##         "up80": [
##             433711072.7506,
##             453885516.7716
##         ],
##         "low80": [
##             320036782.6353,
##             295641687.2737
##         ],
##         "up95": [
##             463798839.8792,
##             495770128.3811
##         ],
##         "low95": [
##             289949015.5067,
##             253757075.6642
##         ]
##     }
## }
## 

ts.analysis uses internally the functions ts.stationary.test,ts.acf,ts.non.seas.decomp,ts.seasonal.decomp, ts.seasonal.model, ts.non.seas.model and ts.forecast. However, these functions can be used independently and depends on the user requirements (see package manual or vignettes).

Time series analysis on OpenBudgets.eu platform

open_spending.ts is designed to estimate and return the autocorrelation parameters, time series model parameters and the forecast parameters of OpenBudgets.eu time series datasets.

The input data must be a JSON link according to the OpenBudgets.eu data model. The user should specify the amount and time variables, future steps to be predicted (default is 1 step forward) and the arima order (if not specified the most appropriate model will be selected according to AIC value).

open_spending.ts estimates and returns the json data (that are described with the OpenBudgets.eu data model), using ts.analysis function.

#example openbudgets.eu time series data
sample.ts.data = 
'{"page":0,
"page_size": 30,
"total_cell_count": 15,
"cell": [],
"status": "ok",
"cells": [{
        "global__fiscalPeriod__28951.notation": "2002",
        "global__amount__0397f.sum": 290501420.64,
        "global__amount__0397f__CZK.sum": 9210928544.2325,
        "_count": 4805
    },
    {
        "global__fiscalPeriod__28951.notation": "2003",
        "global__amount__0397f.sum": 311242291.07,
        "global__amount__0397f__CZK.sum": 9832143974.9013,
        "_count": 4988
    },
    {
        "global__fiscalPeriod__28951.notation": "2004",
        "global__amount__0397f.sum": 5268500701.1,
        "global__amount__0397f__CZK.sum": 170688885714.24,
        "_count": 10055
    },
    {
        "global__fiscalPeriod__28951.notation": "2005",
        "global__amount__0397f.sum": 2542887761.01,
        "global__amount__0397f__CZK.sum": 77204615312.025,
        "_count": 2032
    },
    {
        "global__fiscalPeriod__28951.notation": "2006",
        "global__amount__0397f.sum": 14803951786.68,
        "global__amount__0397f__CZK.sum": 429758720367.32,
        "_count": 13632
    },
    {
        "global__fiscalPeriod__28951.notation": "2007",
        "global__amount__0397f.sum": 16188514346.44,
        "global__amount__0397f__CZK.sum": 445588857385.76,
        "_count": 22798
    },
    {
        "global__fiscalPeriod__28951.notation": "2008",
        "global__amount__0397f.sum": 18231035815.89,
        "global__amount__0397f__CZK.sum": 480643028250.12,
        "_count": 24176
    },
    {
        "global__fiscalPeriod__28951.notation": "2009",
        "global__amount__0397f.sum": 19079541164.68,
        "global__amount__0397f__CZK.sum": 511808691742.54,
        "_count": 26250
    },
    {
        "global__fiscalPeriod__28951.notation": "2010",
        "global__amount__0397f.sum": 22738650575.01,
        "global__amount__0397f__CZK.sum": 597685430364.14,
        "_count": 87667
    },
    {
        "global__fiscalPeriod__28951.notation": "2011",
        "global__amount__0397f.sum": 24961375670.57,
        "global__amount__0397f__CZK.sum": 626230992823.26,
        "_count": 134352
    },
    {
        "global__fiscalPeriod__28951.notation": "2012",
        "global__amount__0397f.sum": 261513607691.41,
        "global__amount__0397f__CZK.sum": 7030666436872.5,
        "_count": 147556
    },
    {
        "global__fiscalPeriod__28951.notation": "2013",
        "global__amount__0397f.sum": 268946402299.09,
        "global__amount__0397f__CZK.sum": 7226220232913.8,
        "_count": 150079
    },
    {
        "global__fiscalPeriod__28951.notation": "2014",
        "global__amount__0397f.sum": 255222816704.9,
        "global__amount__0397f__CZK.sum": 6907598086283.4,
        "_count": 176019
    },
    {
        "global__fiscalPeriod__28951.notation": "2015",
        "global__amount__0397f.sum": 22976062973.62,
        "global__amount__0397f__CZK.sum": 636276111928.46,
        "_count": 213777
    },
    {
        "global__fiscalPeriod__28951.notation": "2016",
        "global__amount__0397f.sum": 12051686541.16,
        "global__amount__0397f__CZK.sum": 325672725401.77,
        "_count": 161797
    }
],
"order": [
    ["global__fiscalPeriod__28951.fiscalPeriod", "asc"]
],
"aggregates": ["", "_count"],
"summary": {
    "global__amount__0397f.sum": 945126777743.27,
    "global__amount__0397f__CZK.sum": 25485085887878
},
"attributes": [""]
}'
 
result = open_spending.ts(
  json_data =  sample.ts.data, 
  time ="global__fiscalPeriod__28951.notation",
  amount = "global__amount__0397f.sum"
  )
# Pretty output using prettify of jsonlite library
jsonlite::prettify(result,indent = 2)
## {
##   "acf.param": {
##     "acf.parameters": {
##       "acf": [
##         1,
##         0.6083,
##         0.1674,
##         -0.1663,
##         -0.1295,
##         -0.0727,
##         -0.0925,
##         -0.1301,
##         -0.1615,
##         -0.1959,
##         -0.2115,
##         -0.1311
##       ],
##       "acf.lag": [
##         0,
##         1,
##         2,
##         3,
##         4,
##         5,
##         6,
##         7,
##         8,
##         9,
##         10,
##         11
##       ],
##       "confidence.interval.up": [
##         0.5061
##       ],
##       "confidence.interval.low": [
##         -0.5061
##       ]
##     },
##     "pacf.parameters": {
##       "pacf": [
##         0.6083,
##         -0.3215,
##         -0.1865,
##         0.25,
##         -0.1593,
##         -0.1764,
##         0.0869,
##         -0.1346,
##         -0.2117,
##         -0.0036,
##         0.0508
##       ],
##       "pacf.lag": [
##         1,
##         2,
##         3,
##         4,
##         5,
##         6,
##         7,
##         8,
##         9,
##         10,
##         11
##       ],
##       "confidence.interval.up": [
##         0.5061
##       ],
##       "confidence.interval.low": [
##         -0.5061
##       ]
##     },
##     "acf.residuals.parameters": {
##       "acf.residuals": [
##         1,
##         0.3097,
##         0.2296,
##         -0.2346,
##         -0.0115,
##         -0.069,
##         -0.0524,
##         -0.0981,
##         -0.0842,
##         -0.1215,
##         -0.0934,
##         -0.0868,
##         -0.0484,
##         -0.2128,
##         -0.115,
##         -0.1051,
##         0.2946
##       ],
##       "acf.residuals.lag": [
##         0,
##         1,
##         2,
##         3,
##         4,
##         5,
##         6,
##         7,
##         8,
##         9,
##         10,
##         11,
##         12,
##         13,
##         14,
##         15,
##         16
##       ],
##       "confidence.interval.up": [
##         0.5061
##       ],
##       "confidence.interval.low": [
##         -0.5061
##       ]
##     },
##     "pacf.residuals.parameters": {
##       "pacf.residuals": [
##         0.3097,
##         0.1479,
##         -0.3857,
##         0.1673,
##         0.0455,
##         -0.2432,
##         0.0379,
##         0.0137,
##         -0.2159,
##         0.0048,
##         0.0175,
##         -0.1445,
##         -0.2757,
##         0.0882,
##         -0.0175,
##         0.2238
##       ],
##       "pacf.residuals.lag": [
##         1,
##         2,
##         3,
##         4,
##         5,
##         6,
##         7,
##         8,
##         9,
##         10,
##         11,
##         12,
##         13,
##         14,
##         15,
##         16
##       ],
##       "confidence.interval.up": [
##         0.5061
##       ],
##       "confidence.interval.low": [
##         -0.5061
##       ]
##     }
##   },
##   "decomposition": {
##     "stl.plot": {
##       "trend": [
##         -823419544.04,
##         1661560665.9804,
##         4624784833.2485,
##         7878983909.6147,
##         9164365784.5264,
##         1249040776.0474,
##         -4351015666.9835,
##         6551641382.3009,
##         57664029716.5692,
##         135646130024.628,
##         199114831577.553,
##         212547970266.68,
##         183231679540.212,
##         110152904453.676,
##         -12061960507.0426
##       ],
##       "conf.interval.up": [
##         100039247758.34,
##         66576136730.9101,
##         60840745923.6514,
##         68328241465.8754,
##         72409579665.7543,
##         65432105297.0944,
##         59676059487.2765,
##         70171989437.8921,
##         121691104870.829,
##         199829194545.675,
##         262360045458.781,
##         272997227822.941,
##         239447640630.614,
##         175067480518.606,
##         88800706795.3375
##       ],
##       "conf.interval.low": [
##         -101686086846.42,
##         -63253015398.9493,
##         -51591176257.1543,
##         -52570273646.646,
##         -54080848096.7016,
##         -62934023744.9996,
##         -68378090821.2435,
##         -57068706673.2904,
##         -6363045437.6908,
##         71463065503.5815,
##         135869617696.325,
##         152098712710.42,
##         127015718449.809,
##         45238328388.7462,
##         -112924627809.423
##       ],
##       "seasonal": {
## 
##       },
##       "remainder": [
##         1113920964.68,
##         -1350318374.9104,
##         643715867.8515,
##         -5336096148.6047,
##         5639586002.1536,
##         14939473570.3926,
##         22582051482.8735,
##         12527899782.3791,
##         -34925379141.5592,
##         -110684754354.058,
##         62398776113.8568,
##         56398432032.4096,
##         71991137164.6884,
##         -87176841480.0559,
##         24113647048.2026
##       ],
##       "time": [
##         2002,
##         2003,
##         2004,
##         2005,
##         2006,
##         2007,
##         2008,
##         2009,
##         2010,
##         2011,
##         2012,
##         2013,
##         2014,
##         2015,
##         2016
##       ]
##     },
##     "stl.general": {
##       "degfr": [
##         5.288
##       ],
##       "degfr.fitted": [
##         4.9747
##       ],
##       "stl.degree": [
##         2
##       ]
##     },
##     "residuals_fitted": {
##       "residuals": [
##         1113920964.68,
##         -1350318374.9104,
##         643715867.8515,
##         -5336096148.6047,
##         5639586002.1536,
##         14939473570.3926,
##         22582051482.8735,
##         12527899782.3791,
##         -34925379141.5592,
##         -110684754354.058,
##         62398776113.8568,
##         56398432032.4096,
##         71991137164.6884,
##         -87176841480.0559,
##         24113647048.2026
##       ],
##       "fitted": [
##         -823419544.04,
##         1661560665.9804,
##         4624784833.2485,
##         7878983909.6147,
##         9164365784.5264,
##         1249040776.0474,
##         -4351015666.9835,
##         6551641382.3009,
##         57664029716.5692,
##         135646130024.628,
##         199114831577.553,
##         212547970266.68,
##         183231679540.212,
##         110152904453.676,
##         -12061960507.0426
##       ],
##       "time": [
##         2002,
##         2003,
##         2004,
##         2005,
##         2006,
##         2007,
##         2008,
##         2009,
##         2010,
##         2011,
##         2012,
##         2013,
##         2014,
##         2015,
##         2016
##       ],
##       "line": [
##         0
##       ]
##     },
##     "compare": {
##       "resid.variance": [
##         2.49022310287957e+021
##       ],
##       "used.obs": [
##         2002,
##         2016,
##         2009,
##         2005.5,
##         2012.5
##       ],
##       "loglik": [
##         -1.7431561720157e+022
##       ],
##       "aic": [
##         3.4863123440314e+022
##       ],
##       "bic": [
##         3.4863123440314e+022
##       ],
##       "gcv": [
##         5.54416871376474e+021
##       ]
##     }
##   },
##   "model.param": {
##     "model": {
##       "arima.order": [
##         2,
##         1,
##         0,
##         0,
##         1,
##         1,
##         0
##       ],
##       "arima.coef": [
##         0.8348,
##         -0.249,
##         -0.9999
##       ],
##       "arima.coef.se": [
##         0.2524,
##         0.2482,
##         0.5954
##       ]
##     },
##     "residuals_fitted": {
##       "residuals": [
##         290501.235,
##         18348491.5673,
##         4388546947.1005,
##         -2696772503.6529,
##         12279728064.2191,
##         1663580465.5181,
##         5162045935.6711,
##         4109968756.8555,
##         6995758466.0538,
##         5772141452.1428,
##         231395392466.55,
##         31316282096.0982,
##         66705686505.6137,
##         -149540611131.324,
##         33819214996.6006
##       ],
##       "fitted": [
##         290210919.405,
##         292893799.5027,
##         879953753.9995,
##         5239660264.6629,
##         2524223722.4609,
##         14524933880.9219,
##         13068989880.2189,
##         14969572407.8245,
##         15742892108.9562,
##         19189234218.4272,
##         30118215224.8599,
##         237630120202.992,
##         188517130199.286,
##         172516674104.944,
##         -21767528455.4406
##       ],
##       "time": [
##         2002,
##         2003,
##         2004,
##         2005,
##         2006,
##         2007,
##         2008,
##         2009,
##         2010,
##         2011,
##         2012,
##         2013,
##         2014,
##         2015,
##         2016
##       ],
##       "line": [
##         0
##       ]
##     },
##     "compare": {
##       "resid.variance": [
##         7.52601888826793e+021
##       ],
##       "variance.coef": [
##         [
##           0.0637,
##           -0.034,
##           -0.0003
##         ],
##         [
##           -0.034,
##           0.0616,
##           -0.0002
##         ],
##         [
##           -0.0003,
##           -0.0002,
##           0.3545
##         ]
##       ],
##       "not.used.obs": [
##         0
##       ],
##       "used.obs": [
##         14
##       ],
##       "loglik": [
##         -371.6686
##       ],
##       "aic": [
##         751.3372
##       ],
##       "bic": [
##         753.8934
##       ],
##       "aicc": [
##         755.7816
##       ]
##     }
##   },
##   "forecasts": {
##     "ts.model": [
##       "ARIMA(2,1,1)"
##     ],
##     "data_year": [
##       2002,
##       2003,
##       2004,
##       2005,
##       2006,
##       2007,
##       2008,
##       2009,
##       2010,
##       2011,
##       2012,
##       2013,
##       2014,
##       2015,
##       2016
##     ],
##     "data": [
##       290501420.64,
##       311242291.07,
##       5268500701.1,
##       2542887761.01,
##       14803951786.68,
##       16188514346.44,
##       18231035815.89,
##       19079541164.68,
##       22738650575.01,
##       24961375670.57,
##       261513607691.41,
##       268946402299.09,
##       255222816704.9,
##       22976062973.62,
##       12051686541.16
##     ],
##     "predict_time": [
##       2017
##     ],
##     "predict_values": [
##       27966100694.5231
##     ],
##     "up80": [
##       142431305172.611
##     ],
##     "low80": [
##       -86499103783.5649
##     ],
##     "up95": [
##       203025524202.564
##     ],
##     "low95": [
##       -147093322813.518
##     ]
##   }
## }
##

News

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("TimeSeries.OBeu")

1.2.3 by Kleanthis Koupidis, 6 months ago


https://github.com/okgreece/TimeSeries.OBeu


Report a bug at https://github.com/okgreece/TimeSeries.OBeu/issues


Browse source code at https://github.com/cran/TimeSeries.OBeu


Authors: Kleanthis Koupidis [aut, cre] , Charalampos Bratsas [aut]


Documentation:   PDF Manual  


GPL-2 | file LICENSE license


Imports devtools, forecast, locfit, jsonlite, stats, trend, tseries

Suggests knitr, rmarkdown


See at CRAN