Least stupid way to forecast a short multivariate time series

I need to forecast the following 4 variables for the 29th unit of time. I have roughly 2 years worth of historical data, where 1 and 14 and 27 are all the same period (or time of year). In the end, I am doing a Oaxaca-Blinder style decomposition on $W$, $wd$, $wc$, and $p$.

time    W               wd              wc               p
1       4.920725        4.684342        4.065288        .5962985
2       4.956172        4.73998         4.092179        .6151785
3       4.85532         4.725982        4.002519        .6028712
4       4.754887        4.674568        3.988028        .5943888
5       4.862039        4.758899        4.045568        .5925704
6       5.039032        4.791101        4.071131        .590314
7       4.612594        4.656253        4.136271        .529247
8       4.722339        4.631588        3.994956        .5801989
9       4.679251        4.647347        3.954906        .5832723
10      4.736177        4.679152        3.974465        .5843731
11      4.738954        4.759482        4.037036        .5868722
12      4.571325        4.707446        4.110281        .556147
13      4.883891        4.750031        4.168203        .602057
14      4.652408        4.703114        4.042872        .6059471
15      4.677363        4.744875        4.232081        .5672519
16      4.695732        4.614248        3.998735        .5838578
17      4.633575        4.6025          3.943488        .5914644
18      4.61025         4.67733         4.066427        .548952
19      4.678374        4.741046        4.060458        .5416393
20      4.48309         4.609238        4.000201        .5372143
21      4.477549        4.583907        3.94821         .5515663
22      4.555191        4.627404        3.93675         .5542806
23      4.508585        4.595927        3.881685        .5572687
24      4.467037        4.619762        3.909551        .5645944
25      4.326283        4.544351        3.877583        .5738906
26      4.672741        4.599463        3.953772        .5769604
27      4.53551         4.506167        3.808779        .5831352
28      4.528004        4.622972        3.90481         .5968299

I believe that $W$ can be approximated by $p\cdot wd + (1 - p)\cdot wc$ plus measurement error, but you can see that $W$ always considerably exceeds that quantity because of waste, approximation error, or theft.

Here are my 2 questions.

1. My first thought was to try vector autoregression on these variables with 1 lag and an exogenous time and period variable, but that seems like a bad idea given how little data I have. Are there any time-series methods that (1) perform better in the face of "micro-numerosity" and (2) would be able to exploit the link between the variables?

2. On the other hand, the moduli of the eigenvalues for the VAR are all less than 1, so I don't think I need to worry about non-stationarity (though the Dickey-Fuller test suggest otherwise). The predictions seem mostly in line with projections from a flexible univariate model with a time trend, except for $W$ and $p$, which are lower. The coefficients on the lags seem mostly reasonable, though they are insignificant for the most part. The linear trend coefficient is significant, as are some of the period dummies. Still, are there any theoretical reasons to prefer this simpler approach over the VAR model?

Full disclosure: I asked a similar question on Statalist with no response.

• Hi there, could you give some more context around the decomposition you wish to do, as I have not seen it applied to time series data? – Michelle Mar 15 '12 at 6:38
• I am breaking up the change into components in the following way: $W^{′}-W=p^{′}∗(w^{′}_{D}-w_{D})+(1-p^{′})∗(w^{′}_{C}-w_{C})+(w_{D}-w_{C})∗(p^{′}-p)+(\epsilon^{′}-\epsilon)$, where primes denote current value of the variables. – Dimitriy V. Masterov Mar 15 '12 at 13:09
• hmmm, how about exclude the outliers first, before regression? – athos Aug 31 '13 at 14:37
• What level of precision do you require? I'm asking because as you know, you can use ARIMA models and get a very low MSE. However, since those models are usually fit using maximum likelihood, it is almost certain that you will overfit. Bayesian models are robust when dealing with little data, but I think you will get a MSE an order of magnitude higher than in ARIMA models. – Robert Smith Sep 29 '14 at 4:56

I understand that this question has been sitting here for years, but still, the following ideas may be useful:

1. If there are links between variables (and the theoretical formula does not work so well), PCA can be used to look for (linear) dependencies in a systematic way. I will show that this works well for the given data in this question.

2. Given there is not much data (112 numbers in total), only a few model parameters can be estimated (e.g. fitting full seasonal effects is not an option), and trying a custom model may make sense.

Here is how I would make a forecast, following these principles:

Step 1. We can use PCA to reveal dependencies in the data. Using R, with the data stored in x:

> library(jvcoords)
> m <- PCA(x)
> m
PCA: mapping p = 4 coordinates to q = 4 coordinates

PC1         PC2          PC3          PC4
standard deviation     0.18609759 0.079351671 0.0305622047 0.0155353709
variance               0.03463231 0.006296688 0.0009340484 0.0002413477
cum. variance fraction 0.82253436 0.972083769 0.9942678731 1.0000000000

This shows that the first two principal components explain 97% of the variance, and using three components covers 99.4% of the variance. Thus, it will be enough to make a model for first two or three PCs. (The data approximately satisfy $$W = 0.234\, wd - 1.152\, wc - 8.842 \,p$$.)

Doing PCA involved finding a $$4\times 4$$ orthogonal matrix. The space of such matrices is 6-dimensional, so we have estimated 6 parameters. (Since we only really use PC1 below, this may be fewer "effective" parameters.)

Step 2. There is a clear trend in PC1:

> t <- 1:28
> plot(m$$y[,1], type = "b", ylab = "PC1") > trend <- lm(m$$y[,1] ~ t)
> abline(trend) I create a copy of the PC scores with this trend removed:

> y2 <- m$y > y2[,1] <- y2[,1] - fitted(trend) Plotting the scores of the other PCs reveal no clear trends, so I leave these unchanged. Since the PC scores are centred, the trend goes through the centre of mass of the PC1 sample and fitting the trend only corresponds to estimating one parameter. Step 3. A pair scatter plot shows no clear structure, so I model the PCs as being independent: > pairs(y2, asp = 1, oma = c(1.7, 1.7, 1.7, 1.7)) Step 4. There is a clear periodicity in PC1, with lag 13 (as suggested by the question). This can be seen in different ways. For example, the lag 13 autocorrelation shows up as being significantly different from 0 in a correlogram: > acf(y2[,1]) (The periodicity is visually more striking when plotting the data together with a shifted copy.) Since we want to keep the number of estimated parameters low, and since the correlogram shows lag 13 as the only lag with a significant contribution, I will model PC1 as $$y^{(1)}_{t+13} = \alpha_{13} y^{(1)}_t + \sigma \varepsilon_{t+13}$$, where the $$\varepsilon_t$$ are independent and standard normally distributed (i.e. this is an AR(13) process with most coefficients fixed to 0). An easy way to estimate $$\alpha_{13}$$ and $$\sigma$$ is using the lm() function: > lag13 <- lm(y2[14:28,1] ~ y2[1:15,1] + 0) > lag13 Call: lm(formula = y2[14:28, 1] ~ y2[1:15, 1] + 0) Coefficients: y2[1:15, 1] 0.6479 > a13 <- coef(lag13) > s13 <- summary(lag13)$sigma

As a plausibility test, I plot the given data (black), together with a random trajectory of my model for PC1 (blue), ranging one year into the future:

t.f <- 29:41
pc1 <- m$y[,1] pc1.f <- (predict(trend, newdata = data.frame(t = t.f)) + a13 * y2[16:28, 1] + rnorm(13, sd = s13)) plot(t, pc1, xlim = range(t, t.f), ylim = range(pc1, pc1.f), type = "b", ylab = "PC1") points(t.f, pc1.f, col = "blue", type = "b") The blue, simulated piece of path looks like a reasonable continuation of the data. The correlograms for PC2 and PC3 show no significant correlations, so I model these components as white noise. PC4 does show correlations, but contributes so little to the total variance that it seem not worth modelling, and I also model this component as white noise. Here we have fitted two more parameters. This brings us to a total of nine parameters in the model (including the PCA), which does not seem absurd when we started with data consisting of 112 numbers. Forecast. We can get a numeric forecast by leaving out the noise (to get the mean) and reversing the PCA: > pc1.f <- predict(trend, newdata = data.frame(t = t.f)) + a13 * y2[16:28, 1] > y.f <- data.frame(PC1 = pc1.f, PC2 = 0, PC3 = 0, PC4 = 0) > x.f <- fromCoords(m, y.f) > rownames(x.f) <- t.f > x.f W wd wc p 29 4.456825 4.582231 3.919151 0.5616497 30 4.407551 4.563510 3.899012 0.5582053 31 4.427701 4.571166 3.907248 0.5596139 32 4.466062 4.585740 3.922927 0.5622955 33 4.327391 4.533055 3.866250 0.5526018 34 4.304330 4.524294 3.856824 0.5509898 35 4.342835 4.538923 3.872562 0.5536814 36 4.297404 4.521663 3.853993 0.5505056 37 4.281638 4.515673 3.847549 0.5494035 38 4.186515 4.479533 3.808671 0.5427540 39 4.377147 4.551959 3.886586 0.5560799 40 4.257569 4.506528 3.837712 0.5477210 41 4.289875 4.518802 3.850916 0.5499793 Uncertainty bands can be obtained either analytically or simply using Monte Carlo: N <- 1000 # number of Monte Carlo samples W.f <- matrix(NA, N, 13) for (i in 1:N) { y.f <- data.frame(PC1 = (predict(trend, newdata = data.frame(t = t.f)) + a13 * y2[16:28, 1] + rnorm(13, sd = s13)), PC2 = rnorm(13, sd = sd(y2[,2])), PC3 = rnorm(13, sd = sd(y2[, 3])), PC4 = rnorm(13, sd = sd(y2[, 4]))) x.f <- fromCoords(m, y.f) W.f[i,] <- x.f[, 1] } bands <- apply(W.f, 2, function(x) quantile(x, c(0.025, 0.15, 0.5, 0.85, 0.975))) plot(t, x$$W, xlim = range(t, t.f), ylim = range(x$$W, bands), type = "b", ylab = "W") for (b in 1:5) { lines(c(28, t.f), c(x$W, bands[b,]), col = "grey")
} The plot shows the actual data for $$W$$, together with 60% (inner three lines) and 95% (outer two lines) uncertainty bands for a forecast using the fitted model.

• Interesting approach. Let me digest this a bit. – Dimitriy V. Masterov Jun 21 at 22:44