---
category: "english"
media: "text"
outlet: "Czech National Bank"
url: "https://www.cnb.cz/export/sites/cnb/cs/menova-politika/.galleries/br_zapisy_z_jednani/2018/download/stanoviska_2018_08.pdf"
date: "2018-12-19"
headline: "Advisor's Opinion on Situation Report No. 8, 2018"
released: "2025-01-31"
genre: "advisor_opinion"
translation: "cnb-stanovisko-poradce-8sz-2018"
body_note: "An English translation of the Czech original, which remains the record of what was written: [Stanovisko poradce k 8. situační zprávě 2018](/komentare/cnb-stanovisko-poradce-8sz-2018/). Part I, the advisor's opinion, is here in full. The original document also carries Part II, prepared by the Financial Stability Department, which is in the [hosted PDF](/komentare/files/cnb-stanoviska-8sz-2018.pdf) and in the document on the CNB site. The four charts stay in the PDF. Numbered footnotes are collected at the end."
---

# Advisor's Opinion on Situation Report No. 8, 2018

The opinion on the Situation Report on economic and monetary developments consists of two parts: (i) an assessment against the price stability objective, written and signed by an advisor to a Bank Board member or, in exceptional cases, by a representative of the Financial Stability Department (Part I), and (ii) an assessment prepared by the Financial Stability Department, which brings in the financial stability objective as well (Part II).

## I. Assessment against the price stability objective

Written by Tomas Havranek.[1]

## I.1 The monetary policy decision

I will start this opinion unconventionally, with a chart:

Chart 1. The model interest rate is rising fast, client rates for consumers are not

The chart compares PRIBOR, to which our model consumer in g3 responds, with the effective client rate the average consumer actually faces (I explain the construction of the effective rate in section I.3). The model does not see client rates. If it saw how slowly they respond to changes in our repo rate, it would imply a faster increase in rates toward normal levels. Consumption makes up roughly half of aggregate demand, so without an increase in the effective interest rate on consumption, domestic inflationary pressures are not being damped as they should be. A further technical argument for continuing to raise rates is that the relationship between consumption and interest rates (even in effective terms) is apparently weaker than assumed by our model, and indeed by most other DSGE models. I also discuss this empirical relationship in more detail in section I.3.

But even leaving both of the above facts aside, the current forecast is consistent with a further increase in interest rates. Despite the halt in year-on-year food price growth in November, the upside risks to inflation remain strong (above all the exchange rate, see section I.2) and were communicated in advance through the sensitivity scenario. The wage bill is still growing at a pace of around 10%, the average wage by 6% in real terms, while labor productivity growth is less than one quarter of that figure (p. 23 of the Report). And historically, wage growth has not slowed much without a recession or a substantial tightening of monetary policy. It is thus a matter of time before a larger share of firms operating in the Czech market start to resist the squeeze on their margins by raising prices faster. The right response of forward-looking monetary policy here is to moderate inflationary pressures and raise the repo rate by 25 basis points. I recommend raising the discount and Lombard rates by the same amount.

By contrast, I do not recommend raising rates by 50 basis points. While, given the above, I believe that markedly higher rates would suit our economy at this stage of the cycle, moves larger than the standard 25 basis points create an impression of urgency in central banking. We are not in an urgent situation, mainly because we have resolved, as the only central bank in the advanced world today, to raise rates at consecutive meetings. Today the weak koruna gives us an opportunity to continue with rate normalization, because neither the exchange rate nor client rates are delivering enough tightening of monetary conditions yet. A fifth consecutive increase looks dramatic, but in historical context it is common during an expansion. The US Fed, for example, raised rates seventeen times in a row between 2004 and 2006.

## I.2 Assessment of the message of the Situation Report and the forecast

Forecasting the exchange rate is hard (many econometrician colleagues would say outright impossible). So we cannot demand precise forecasts from any model, and errors of a few percent even in the very next quarter should not surprise anyone. Besides, in our model the point is not primarily forecasting but the consistency of the exchange rate with other variables, above all the implied interest rate path.

What we should want from a model, though, is that its errors not be systematic. Our g3 forecast errors for the exchange rate are systematic, as the following chart shows. In it I plot the forecast from each Situation Report for the following quarter and compare it with the actual average for that quarter (the actual exchange rate for the first quarter of 2019 is thus naturally missing from the chart):

Chart 2. Our model systematically forecasts the koruna too strong against the euro in the following quarter

In the chart I consider only the forecasts published after the end of the exchange rate commitment, all of which erred in the same direction, by half a koruna on average. The pre-exit forecasts, by contrast, assumed that the end of the commitment would come later than April 2017, and therefore had the exchange rate for the following quarter fixed at CZK 27 to the euro. Since the exit came earlier, those few forecasts briefly understated the koruna's strength, for technical reasons. But this has nothing to do with the model's performance, good or bad, as an exchange rate predictor, and it does not stand up as an argument for the symmetry of the model.

This asymmetry matters for the policy recommendations we build on the forecasts. By our rule of thumb, an average error of CZK 0.5 in the next quarter (propagated onward) means that at the time of the decision on each Situation Report there was, on average, room for two more “hikes” beyond what the forecast then recommended.

That room exists now as well. What matters for today's decision is that the forecast assumes an average exchange rate of CZK 25.1 to the euro in the first quarter of 2019. Given market developments, however, such an outcome looks unlikely. And the distortion associated with the koruna weakening at the end of the year is not very pronounced this time. That is partly thanks to the activity of the National Bank of Slovakia and the Bank of Israel, which exploit the arbitrage, intervene in FX swaps, and indirectly help the koruna appreciate. In January, then, we cannot expect the koruna to strengthen as the end-of-year effect expires. If anything, the sensitivity scenario is materializing, and it calls for a rate increase of as much as 50 basis points. In any case, with one standard “hike” we cannot go wrong.

## I.3 Thematic comment on the Situation Report

Household consumption is the most important component of aggregate demand through which monetary policy affects inflation. In our g3 model, as in most DSGE models used in central banks (unlike some models with heterogeneous agents, e.g. Kaplan et al., 2018), monetary policy acts on consumption mainly through intertemporal substitution. Put simply: when real rates rise, consumers will want to save more and borrow less. The parameter that governs the relationship between rates and consumption in the model is the elasticity of intertemporal substitution. It is the crucial parameter, and it drives the main outputs of DSGE models:

Chart 3. The elasticity of substitution (EIS) determines the model response of consumption to a change in rates

The larger the elasticity, the larger and faster the response of consumption to rate changes. For technical reasons, the simulation in the chart above does not use the g3 model but the world's best-known DSGE model, Smets and Wouters (2007), which is built for a closed economy. The effect of the elasticity is presumably similar in our case, but this cannot be verified directly, because the g3 model uses a logarithmic utility function (in which the elasticity cannot be changed; it is fixed at 1). The simplification of assuming a logarithmic utility function is quite common around the world, much like the assumption of a Cobb-Douglas production function.

A credible estimate of this elasticity (using micro data and satisfactory identification) is unfortunately not available for the Czech Republic, owing to persistent data limitations. In Havranek (2015) I therefore summarize all available estimates for other countries: 2,735 coefficients published in 169 studies. After correcting for various biases, I obtain a mean elasticity of 1/3, and values close to 1 are exceptional in the literature. In Havranek et al. (2015) my colleagues and I then show that countries with higher household stock market participation have a higher elasticity of intertemporal substitution, since households exposed to the stock market are generally more financially literate and pay more attention to interest rate changes.

All in all, then, it seems unlikely that the elasticity of substitution for the Czech Republic exceeds 1/3. If it equals 1/3, the short-run response of consumption to rate changes is roughly one third of what g3 assumes. Again this is a simplification, because the Euler equation in question involves expectations of consumption growth and expectations of the real rate. Still, the comparison of the turquoise and green lines in Chart 3 clearly shows that the difference is a significant one. Consumption may respond to rate changes considerably less than our model assumes.

Another question about the model transmission is the definition of the word “rates”. Our model has the 3M PRIBOR in the consumption equation, but a consumer can neither borrow nor deposit anything at that rate. Households face client rates, to which they then respond directly through intertemporal substitution or, as Kaplan et al. (2018) stress, indirectly through changes in income. Unlike PRIBOR, however, client rates on total deposits and on consumer loans have been falling since the exit:

Chart 4. The client rates most relevant for consumption have been falling since the exit

Both indicators are close to their all-time lows. Should we also count mortgage interest among the rates relevant for consumption? Mortgage rates are rising, which may change the picture (though they are also rising on account of the tightening of macroprudential policy, not just monetary policy). I lean toward yes, since housing costs can in many cases be regarded as consumption, as Mojmir Hampl has repeatedly argued. The size of the mortgage payment affects household consumption behavior, as many of us observe from our own experience.

As an illustration, I can build a rough indicator of the effective rate faced by broadly defined consumption, that is, including home purchases. I simply weight the rates on aggregate household savings (which are the opportunity cost of consumption), on consumer loans, and on mortgages by the volume of each category. The result is the blue line in Chart 1, with which this opinion opened. The rise in mortgage rates since the exit has outweighed the decline in the other two categories. Even so, the cumulative rise in the effective rate for consumption (which the model does not see) is just 0.1 percentage point, while for PRIBOR (which the model works with) it was 1.5 percentage points. By this logic, then, the rate increases so far have had a negligible effect on consumption and will go on having a negligible one in the months ahead. (A recent internal analysis by Michal Franta of errors in our consumption forecasts also points out that our forecasts have lately tended to underestimate long-run consumption growth, although the problem should show up fully only at the transmission horizon.) If we assume that the rest of the transmission works flawlessly, the overall transmission mechanism is weakened by almost one half, since consumption makes up half of aggregate demand. Put simply, it is as if since the exit we had delivered only just under four “hikes” instead of seven.

To conclude: the available data on client rates end in October 2018. For November and December, anecdotal evidence already suggests movement in mortgages and savings accounts (see also p. 28 of the Report), which raises the effective rate for consumption. So it will not do to say that transmission to consumption does not work: it was merely strongly muted, or even halted, in the post-exit period. Examining the reasons for this muting is beyond the scope of a short opinion. What matters is that at least three post-exit “hikes” have as a result fizzled out without much effect, without the model taking this into account. This fizzling may soon show up in higher inflation relative to the forecast, given that the time at which these hikes should have their greatest effect on inflation is approaching. To keep inflation at the target, it will therefore be necessary to raise rates at every meeting until wage growth slows tangibly.

## I.4 Specific comments and questions

- The eighth Situation Report is exceptionally well written and clear.
- This opinion is not a critique of the g3 model; it ranks among the best DSGE models for open economies in the world, and we are rightly proud of it. It was calibrated as a whole so that its impulse responses match reality. But shouldn't the individual parameters have an economic interpretation? If they do not, and the calibrated elasticity of intertemporal substitution, for example, does not match observed consumer behavior, does the model really have microeconomic foundations? Was it not fully calibrated back before the crisis? And might the economy's responses today therefore differ from the model's dynamics? Of course, kicking DSGE models is fashionable in macroeconomics these days. I, by contrast, think they are a useful tool. Lately, though, its limitations have been showing more too.
- One limitation described in this opinion is the exchange rate forecast. Based on the arguments in section I.2 above, the yellow triangle for the exchange rate in the GRIP risk chart of this small Situation Report should lie roughly in the middle of the upper right quadrant, not near the origin. I propose extending the expert adjustments to the entire forecast horizon, so that the exchange rate forecast is driven less by the model dynamics and more by information from the market. As the persistently low rates on household savings show, arbitrage between savings accounts and the repo rate does not work fully even in the domestic market. No wonder, then, that the arbitrage in the FX market through the UIP condition assumed by the model does not work fully either. In any case, the consensus in financial econometrics is that few models can beat a random walk at forecasting the exchange rate.
- Another limitation is that g3 does not work with client rates in the consumption equation. It is probably not feasible to replace PRIBOR with another rate there, even though PRIBOR is not very relevant for consumption (see Chart 1 at the beginning of this opinion). But could we build a small satellite model, perhaps just by tying the key variables together with identities, that would work with client rates (and perhaps also with bond yields for the FX market, which, incidentally, are below the 3M PRIBOR even at considerably longer maturities, see p. 27 of the Report) and in future provide a “sanity check” for our core DSGE model?
- Chapter III.6.1 on transmission reads too optimistically to me. Only the last sentence touches on the problems in transmission (which, after all, affect half of aggregate demand).
- Average wage growth in the public sector keeps surprising us on the upside; the latest figure is 12% year on year. The forecast says this pace will moderate to 4% by 2020 (p. 22). What is the logic of this moderation? Parliamentary elections are due in at most 2.8 years, and the entire horizon of our fiscal forecast (p. 26) shows that “the resources will be there”. True, the model works primarily with wages in market sectors, but higher disposable income will raise inflation even if its source is public money.
- The strength of current wage pressures is better conveyed by the growth of the wage bill and of the median wage (both growing at close to 10%; p. 22) than by the growth of the average wage. Employment is still rising, and firms are hiring mostly employees paid below the average, which holds measured average wage growth down. Once employment growth stops, the average wage will grow faster for purely statistical reasons.
- We already have monthly data from the Land Registry on transaction prices of houses and apartments. The October figures suggested a return to double-digit annual growth, so this series could also help us in the argumentation around the approval of the amendment to the CNB Act. When will it be possible to construct the CPIH index on a monthly basis? Can we make it in time for the next big Situation Report, where we publish this index?

## I.5 References

- Havranek, T. (2015): [“Measuring Intertemporal Substitution: The Importance of Method Choices and Selective Reporting,”](/eis/) Journal of the European Economic Association 13(6), 1180-1204.
- Havranek, T. & Horvath, R. & Irsova, Z. & Rusnak, M. (2015): [“Cross-Country Heterogeneity in Intertemporal Substitution,”](/substitution/) Journal of International Economics 96(1), 100-118.
- Kaplan, G. & Moll, B. & Violante, G. (2018): “Monetary Policy According to HANK,” American Economic Review 108(3), 697-743.
- Smets, F. & Wouters, R. (2007): “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach,” American Economic Review 97(3), 586-606.

## Notes

[1] I thank Michal Franta, Ales Michl, and Jiri Schwarz for discussion. The views here are mine alone.