Since last time I had a close look at Lithuania not a whole lot of new data has been published. In this way, I can't make a comprehensive update most notably because the full range of data I am tracking has not been released for Q4 2007. However, among the pieces of data which actually have been released a couple of noteworthy things stand out. Firstly, the provisional estimate for Q4 GDP (q-o-q) was confirmed to have shown a contraction. This came on the back of a dubiously strong Q3 so I am not sure how much we can read into this in terms of determining the question on everybody's lip as to whether it will be a hard or soft landing. In this light, the y-o-y figures still look quite strong.
More generally, we have also seen that inflation measured by the EU HICP methodology continued to rise on a monthly (y-o-y) basis ticking in into double digit territory at 10 % in January. That is not the way we would like to see things moving. Bloomberg furthermore alerts us to the fact that wages continued their brisk growth rate in Q4 posting a 18.5% increase on an annual basis. All this is old news really but what has been persistently missing in the general discourse is an explanation of why exactly we have seen this very rapid rate of wage growth and inflation across the Eastern European edifice.
In my analyses I have emphasised how the exclusion of demographics as a variable grossly tend to oversimplify the situation. Now it seems that somebody somewhere is beginning to re-assess the way they look at this. I was consequently (positively) surprised to read the following Bloomberg snippet on the wage growth release from Lithuania ...Lithuanian wages grew an annual 18.5 percent in the fourth quarter as accelerating inflation and lack of available labor pressured employers to raise salaries. The average nominal monthly wage rose to 2,052 litai ($880), the Vilnius-based statistics office said on its Web site today. Wages grew 17.9 percent in the previous quarter. Lithuania's falling unemployment rate and labor migration to Ireland and the U.K. have fueled wage increases and added to inflation. Wages, which are advancing faster than productivity, are hurting the country's competitiveness and may hamper exports, banks including Goldman Sachs have said. So, Goldman et al. as well as the rest of us can all see the non-sustainability in this from the point of view of the fundamental relationship between wages/inflation, productivity and competitiveness. But this is merely the effect. What is much more important is the cause and it is pretty clear I think that the demographic dynamics need to be considered too. More generally, we could also invoke the point that almost all Eastern European countries saw their fertility rates collapse on the back of the Berlin Wall's fall and subsequent opening up towards the West. However, let us stay with migration for now. The graphs below show in pictures the point Bloomberg (Milda Seputyte) makes;
If we scrutinize the two first graphs it does not seem to be a coincidence that the beginning of 2005 was when the tightening of the labour market really took off. If we assert, as can be observed from the subsequent graphs fielded below, that the inflows of credit and investment really shot up in the middle of 2003 and if we build in the point that the labour market tends to lag the business cycle it seems perfect in line with economic theory. In Lithuania's and Eastern Europe's case however especially the pace of tightening and as a derivative inflation have been staggering. One of the key reasons for this has to be found, in my opinion, in the third graph which shows the net annual outflow of migration. We note immediately that the overall number of emigrants has declined substantially over an 11 year period with a clear break in 00-01. Yet if we couple this with the collapse of fertility seen in the beginning of the 1990s we get a blueprint for, at least a part of, the very rapid run on capacity in the Eastern European/Lithuanian economy.In this last part of this note I am going to shift gears a little bit. In my latest note on Lithuania I fielded graphs on Lithuania's net international investment position (NIIP) which measures the difference between a country's assets and liabilities and thus acts as proxy for the financing/composition of the external balance. Below I reproduce the graph ..
One of the striking features of the situation in Lithuania and the rest of the Baltics is the extent to which bank loans have been used to finance the external deficit. Regular readers will know this story very well as it tells about how foreign banks have been very eager to supply loans denominated in Euros and Swiss Francs to an extensive part of households and corporations in the CEE and Baltic economies. We can all see and interpret the graphs above but what follows will be a more formal and, according to some, rigorous derivation of this relationship. As such beware since math and econometrics will now follow. In order to show this three models are estimated below. All models are estimated in first differences in order to correct for non-stationarity in the time series. Autocorrelation and heteroscedasticity have been deemed not to exist in the models on the basis of rudimentary graphical analysis. As such, we are dealing with three pretty dodgy models which nonetheless convey the message needed. The data is quarterly in LTL million from Q2 1997 to q3 2007.
The first model regresses the change in the NIIP on the change in the inflows of FDI, portfolio investments(PI) and bank loans(BL).
F'(NIIP) = -592.1B0 - 0.471F'(FDI) + 0.154F'(PI) - 0.353F'(BL) + ut
(t: -3.6) (t: -4,39) (t: -0.66) (t: -4.64)
The signs for all variables are in accordance with theory except for portfolio investments. However, this variable is not statistically significant at 5% so for all intent and purposed it could be excluded from the model over the sample period in question. All other variables are statistically significant at 5% with some margin. The model itself has a very high explanatory power which is quite as expected. The R-square is 0.534 with a subsequent F value of 14.5 which is significantly above the critical value in question. What I particularly take away from this model is that the coefficient estimated for bank loans is almost as high as the corresponding coefficient for FDI which is rather unusual I would say even if I don't have empirical evidence readily at hand. This thus confirms, in numbers, the relatively high importance of foreign credit in the Eastern European expansion assuming that the Lithuanian case mirrors the general CEE situation. The confirmation of this point leads to the estimation of the second model which regresses the change in the NIIP exclusively on the change in the inflow of bank loans.
F'(NIIP) = -896.1B0 - 0.365F'(BL) + ut
(t: -5.76) (t: -4.12)
At a first glance the second model does not seem to add much to the general picture. Both the intercept and the estimated coefficient of BL are statistically significant at 5 % with the latter almost equalling the value of model 1. However, if we look at the reported R-square of 0.2975 with a corresponding F value of 16.92 (highly significant at 5%) it conveys, quite remarkably I think, the point that especially foreign credit has been (and is) tantamount in Lithuania's expansion.
The last model or models involves the econometric technique of testing for structural breaks in relationships over time. And by all means, if you did not like what preceded this you won't like the following either. As can be observed from visual inspection of the figures above there seems to be a shift somewhere around Q2 2003. Such structural breaks if present can be shown with statistical tests. In the appendix below I reproduce the full derivation of the so-called Chow test. What suffice to know at this point are the main results. If we thus split up the data set from q2(1997) to q2(2003) and from Q3(2003) to Q3(2007) and test for a structural break in the model originally estimated with all variables we obtain an F value of 6.69 Under the H0 of parameter stability (no structural break) we clearly reject this hypothesis and thus conclude that a structural break is present both at 5% (against 2.69) and 1% (against 4.02)
I had a couple of objectives in this post. First of all I wanted to bring you up to speed with the latest news on Lithuania even if the whole range of data from Q4 is not yet available. Given the fact that GDP seems to be trending down at the same time as the rate of inflation keeps on rising confirms my general pessimist stance towards the soft v hard landing debate. I also noted with positive surprise that the main discourse now seems to be factoring in the region's demographics (in the specific case Lithuania's) in explaining the labour market and wage situation. Lastly, I ventured into the realms of econometrics and tried to formalize the relationship between the rolling financing of Lithuania's external balance and the inflows of bank loans.
The formal expression for the F value used to evaluate the hypothesis of a structural break is as follows; RSS is residual(error) sum of squares.
F = [(RSS(r) - RSS(ur))/k] / [(RSS(ur))/(n1+n2-2k)]
The F value in this case follows a F distribution with [k,(n1+n2-2k)] degrees of freedom. K is the number of explanatory variables including the intercept in the model (4 in this case) and n marks the sample sizes. RSS(r) is estimated by the original regression over the entire sample period and RSS(ur) is estimated from the regressions from the two sub periods. As such, RSS(ur) equals RSS1+RSS2 where RSS1 is the estimated RSS from sub period 1 and vice versa for RSS2. The Chow test formally evaluates the H0 that a1 = d1 = y1 against the Ha that a1 is not equal to d1 and y1 (sorry for lack of adequate notation). In the present case we are thus testing the following joint hypothesis. The subscript 1 and 2 denotes the sub period to which the estimated parameter belongs.
H0: [bo= (b0)1= (b0)2], [F'(FDI) = F'(FDI)1 = F'(FDI)2], [F'(PI) = F'(PI)1 = F'(PI)2] , and [F'(BL) = F'(BL)1= F'(BL)2]
Ha: Essentially testing against the equality signs above.
In our context it leads to the following calculation.
F= [62625030-(4598233+8730034)/4] / [(62625030) / (25+17-2(4))] = 6.69
F~ [4, 34] = 2.69 (5%)
F~ [4, 34] = 4.02 (1%)
As 6.69 > both 2.69 and 4.02 we reject the Ho of no structural break.