BThe Normative Benchmark
This appendix develops the benchmarks summarized in Section 4. I use the benchmark to calculate two empirical objects. First, I want a country-specific optimal equity share \(\alpha^{*}_{c}\) that I can compare the AI financial advice to. Second, I want to be able to calculate a calibration slope that measures how much of the cross-country variation that should appear in the advice actually does appear. This slope nests the “U.S.-default” hypothesis (H2), where the slope would be zero. To this end, I use a simple Merton benchmark that is deliberately static and closed form. While a full life-cycle model in the tradition of Cocco et al. (2005) would add realism on several margins, it would also come at the cost of a large calibration apparatus (for such a model, see, e.g., Choukhmane et al. 2026). Instead, I use a simpler model to illustrate the central mechanism, like background wealth crowding equity into the liquid portfolio.
B.1Benchmark S: the stake as a standalone portfolio
Consider the investor described in the prompt: age \(a=40\), liquid savings \(W\) (the local-currency equivalent of $50,000), choosing how to split \(W\) between a broad equity index fund and domestic government bonds. Let \(r_{c}\) denote the real return on the safe asset in country \(c\), let the expected excess return on the fund be \(\mu_{c}-r_{c}\) with standard deviation \(\sigma_{c}\), both measured in local currency, and let the investor have constant relative risk aversion over terminal wealth with coefficient \(\gamma_{c}\). This setup gives the Merton (1969) and Samuelson (1969) rule for the optimal equity share, equation (1) in Section 4.1. The optimal equity share does not depend on the level of wealth. If returns are independent over time, the horizon also does not enter the equation. Because the prompt restricts the answer to a fraction between zero and one, every benchmark below is truncated to the unit interval, \(\alpha^{*}_{c}=[\theta^{*}_{c}]_{0}^{1}\) with \([x]_{0}^{1} \equiv \min\{1,\max\{0,x\}\}\).
The first benchmark takes the prompt at face value: the investor is asking about this $50,000 as a self-contained decision (the investor is “looking to invest $50,000 over a five-year horizon”).
Under Benchmark S the optimal share is simply equation (1), \[\alpha^{S}_{c} \;=\; \left[\frac{\mu_{c}-r_{c}}{\gamma_{c}\,\sigma_{c}^{2}}\right]_{0}^{1}. \tag{8}\]
How can we measure \(\mu_{c}\) and \(\sigma_{c}\) in this context? The prompt says “a broad stock market index fund,” which could mean the local equity market or a global fund held from country \(c\). I take the world equity index measured in local currency as the baseline. This matches the instruments named in Version C (MSCI World trackers in most countries) and avoids attributing “bias” to any sensible reluctance to recommend concentrated exposure to a small home market. I also do a robustness check with \(\mu_{c}\) and \(\sigma_{c}\) based on the local stock market index in Table A4.
Under the world-index reading, the expected excess return is common across currencies by uncovered interest parity. I set the expected excess return at approximately 5.3 percent, the arithmetic mean of the MSCI World index (net, in U.S. dollars, 1999–2025) over the U.S. long-term rate, both nominal. I also use a common risk aversion coefficient for each country, \(\gamma_{c}=\bar{\gamma}\). This assumption can be changed, but note that risk aversion on the country level does not track the country coefficient.
These assumptions imply that cross-country variation in \(\alpha^{S}_{c}\) comes from local-currency volatility alone. I compute \(\sigma_{c}\) as the standard deviation of annual nominal local-currency returns on the index, converting the U.S.-dollar return with the annual exchange rate of the Global Macro Database (Müller et al. 2025). The baseline uses these unhedged moments, but Table A4 also reports a currency-hedged variant.
I report results on a risk-aversion grid \(\bar{\gamma}\in\{2,3,5,10\}\) with baseline \(\bar{\gamma}=3\). At the baseline, equation (8) produces interior optimal shares from about 0.21 in Turkey to 0.53 in South Korea. Choukhmane et al. (2026) estimate by simulated method of moments a coefficient of relative risk aversion of about 5 that rationalizes the advice of current-generation models, which corresponds to the \(\gamma=5\) point of the sensitivity grid, one step above the baseline. At \(\gamma=2\) the U.S. benchmark rises to about 0.78, and at \(\gamma=10\) shares compress to between 6 and 16 percent across countries, implausibly low relative to observed advice.
Two choices above widen the cross-country spread. First, I hold the arithmetic premium common across currencies. Uncovered interest parity holds in logs, so a common log premium would instead raise the arithmetic premium in volatile currencies by about \(\tfrac{1}{2}(\sigma_{c}^{2}-\sigma_{US}^{2})\), with \(\sigma_{c}\) the volatility of log returns, and with it the benchmark share. Second, I use nominal rather than real volatility. For an investor whose prices move with the exchange rate, nominal volatility overstates currency risk. In the data this matters mainly for Turkey. Both choices lower the benchmark share where currencies are volatile, so the prescribed deviations for these countries are, if anything, overstated. For the U.S. cell the equity volatility is 18.4 percent (Appendix Table A6), which implies a share of 0.52 at the baseline \(\bar{\gamma}=3\).
B.2Benchmark I: adding pension wealth and human capital
The prompt also says “I have a standard pension plan otherwise,” which invites the model to consider the investor’s full balance sheet. The second benchmark therefore integrates two non-tradable assets that a 40-year-old holds outside the liquid portfolio: pension wealth \(P_{c}\), the present value of future pension benefits, and human capital \(H_{c}\), the present value of future labor income (Bodie et al. 1992; Campbell and Viceira 2002; Gálvez and Paz-Pardo 2026). If a fraction \(\zeta\) of human capital and a fraction \(\omega\) of pension wealth are equity-like, equation (2) with \(B^{s}_{c}=(1-\zeta)H_{c}+(1-\omega)P_{c}\) and \(B^{r}_{c}=\zeta H_{c}+\omega P_{c}\) gives the optimal share of the liquid portfolio, \[\alpha^{I}_{c} \;=\; \left[\theta^{*}_{c}\,\frac{W + H_{c} + P_{c}}{W} \;-\; \frac{\zeta\,H_{c}+\omega\,P_{c}}{W}\right]_{0}^{1}. \tag{9}\] With \(\zeta=\omega=0\) this is the rule of Merton (1971) for an investor with riskless non-tradable income, in the form restated by Cocco et al. (2005, equation 12). Safe outside wealth makes the investor less exposed to her liquid portfolio, so she should hold more equity in it. The more generous the pension system, the stronger this push. The second term works in the opposite direction when labor income co-moves with the stock market or when the pension itself is invested in stocks. The equity-like parts of human capital and pension wealth are treated as earning the market return and both are discounted at the safe rate. The baseline sets \(\omega=0\), a reasonable description of public and defined-benefit pillars, although treating pension claims as bond-like is itself contestable for pay-as-you-go systems whose benefits are indexed to wages. The exploratory probes of Section 6.3 state the pension’s composition directly, which sets \(\omega=0\) for a guaranteed annuity and \(\omega=1\) for a pension invested in stocks.
I calibrate both assets for the average earner. Human capital \(H_{c}\) is the present value of the average wage \(\bar{w}_{c}\), growing at the real rate \(g_{c}\), until the retirement age \(R_{c}\). Pension wealth \(P_{c}\) is the present value of a pension equal to the net replacement rate \(\rho_{c}\) times the final wage, paid for the \(L_{c}\) expected years of retirement. Both are discounted at the local real safe rate. The notes to Appendix Table A6 give the formulas and sources.
With these inputs, \(H_{c}+P_{c}\) for an average 40-year-old ranges from three times \(W\) in India to more than eighty times in Switzerland and Turkey, so equation (9) puts \(\alpha^{I}_{c}\) at the corner of 100 percent equity in essentially every country. At \(\zeta=0\) the corner holds for all 21 countries through \(\bar{\gamma}=5\) and breaks at \(\bar{\gamma}=10\) only for four countries (Brazil, India, Mexico, South Africa). At \(\zeta=0.2\) it holds everywhere through \(\bar{\gamma}=3\). Benchmark I therefore cannot rationalize cross-country variation in the advice, and interior advice already rules it out in levels. I use it instead for its comparative statics, which give the predicted signs for the tests of Section 6. The uncapped share inside the brackets rises with a generous pension, falls when the pension is invested in stocks, and falls with age as human capital declines.
B.3Taking the benchmark to the data
The benchmark is computed from external country-level inputs and then confronted with the experimental estimates using two measures, the advice gap and the calibration slope. Appendix Table A6 reports every country-specific input, with sources in the table notes.
Advice gaps.
For each country, I compute the gap between mean LLM advice and the benchmark, \[\Delta_{c} \;=\; \bar{y}_{c} - \alpha^{S}_{c}, \tag{10}\] where \(\bar{y}_{c}\) is the mean recommended fraction in the primary cell (Version B, Q1, age 40). The gap \(\Delta_{c}\) is a descriptive level comparison and carries no confidence interval.
The calibration slope.
The next measure examines whether cross-country movements in the benchmark show up in the advice. Let \(\hat{b}_{c}\) denote the estimated country fixed effect from equation (3) (the LLM advice deviation from the U.S. baseline) and let \(\delta^{*}_{c}=\alpha^{S}_{c}-\alpha^{S}_{US}\) denote the benchmark’s prescribed deviation. The regression \[\hat{b}_{c} \;=\; \kappa_{0} + \kappa\,\delta^{*}_{c} + e_{c} \tag{11}\] summarizes calibration in a single number. Full normative calibration implies \(\kappa=1\): advice moves one for one with the benchmark. The U.S.-default hypothesis implies \(\kappa=0\). Note that the converse does not hold, since large country effects orthogonal to fundamentals also produce a zero slope. I therefore report the gap table and the slope together. Intermediate values measure partial calibration. The registration specifies this regression and its inference but leaves it at the descriptive altitude. Inference follows the rest of the paper: a wild bootstrap over the twenty non-U.S. countries (Webb weights, 9,999 replications) as the primary procedure with HC2 standard errors alongside, a two-stage bootstrap that propagates the estimation error in \(\hat{b}_{c}\), and DFBETA diagnostics pre-specified for the seven countries with thin benchmark data.
B.4Advice gaps and the calibration slope
The benchmark prescribes large differences in advice across this panel. At the baseline grid point (\(\bar{\gamma}=3\)), the standalone benchmark \(\alpha^{S}_{c}\) ranges from 0.21 in Turkey and 0.23 in Brazil, whose local-currency volatilities are the highest in the panel, to 0.53 in South Korea and 0.52 in the United States, a prescribed spread of thirty-two percentage points. Mean advice spans twelve points, from 0.46 to 0.58 (Table A4). The advice range is thus about a third of the prescribed range, but the two ranges are not aligned. Turkey, whose prescribed share is the lowest in the panel, receives mean advice of 0.54, in the middle of the realized band and 3.1 points below the U.S. baseline, a difference that does not survive the Holm correction (Table A8). The calibration slope below measures the alignment directly.
| Mean | Unhedged | Hedged | Home index | ||||
|---|---|---|---|---|---|---|---|
| Country | advice | \(\alpha^{S}_{c}\) | Gap \(\Delta_{c}\) | \(\alpha^{S}_{c}\) | Gap | \(\alpha^{S}_{c}\) | Gap |
| Turkey | \(0.54\) | \(0.21\) | \(0.33\) | \(0.52\) | \(0.02\) | \(0.21\) | \(0.33\) |
| Brazil | \(0.46\) | \(0.23\) | \(0.23\) | \(0.52\) | \(-0.06\) | \(0.38\) | \(0.08\) |
| Mexico | \(0.54\) | \(0.38\) | \(0.17\) | \(0.52\) | \(0.02\) | — | — |
| Japan | \(0.58\) | \(0.43\) | \(0.15\) | \(0.52\) | \(0.06\) | \(0.51\) | \(0.07\) |
| Poland | \(0.50\) | \(0.35\) | \(0.15\) | \(0.52\) | \(-0.03\) | \(0.14\) | \(0.35\) |
| United Kingdom | \(0.58\) | \(0.49\) | \(0.09\) | \(0.52\) | \(0.05\) | \(0.57\) | \(0.00\) |
| Sweden | \(0.57\) | \(0.48\) | \(0.09\) | \(0.52\) | \(0.05\) | \(0.44\) | \(0.13\) |
| India | \(0.55\) | \(0.46\) | \(0.09\) | \(0.52\) | \(0.03\) | \(0.38\) | \(0.17\) |
| Australia | \(0.58\) | \(0.49\) | \(0.08\) | \(0.52\) | \(0.05\) | \(0.74\) | \(-0.17\) |
| South Africa | \(0.56\) | \(0.49\) | \(0.07\) | \(0.52\) | \(0.04\) | \(0.58\) | \(-0.02\) |
| South Korea | \(0.58\) | \(0.53\) | \(0.06\) | \(0.52\) | \(0.06\) | \(0.42\) | \(0.16\) |
| United States | \(0.57\) | \(0.52\) | \(0.05\) | \(0.52\) | \(0.05\) | \(0.68\) | \(-0.11\) |
| China | \(0.52\) | \(0.48\) | \(0.04\) | \(0.52\) | \(0.00\) | — | — |
| Canada | \(0.56\) | \(0.52\) | \(0.04\) | \(0.52\) | \(0.04\) | \(0.68\) | \(-0.12\) |
| Finland | \(0.55\) | \(0.52\) | \(0.03\) | \(0.52\) | \(0.02\) | \(0.08\) | \(0.47\) |
| Switzerland | \(0.53\) | \(0.51\) | \(0.02\) | \(0.52\) | \(0.00\) | \(0.46\) | \(0.06\) |
| Germany | \(0.53\) | \(0.52\) | \(0.01\) | \(0.52\) | \(0.01\) | \(0.35\) | \(0.18\) |
| Netherlands | \(0.53\) | \(0.52\) | \(0.01\) | \(0.52\) | \(0.01\) | \(0.43\) | \(0.09\) |
| Spain | \(0.53\) | \(0.52\) | \(0.01\) | \(0.52\) | \(0.00\) | \(0.34\) | \(0.19\) |
| France | \(0.52\) | \(0.52\) | \(-0.00\) | \(0.52\) | \(-0.01\) | \(0.31\) | \(0.20\) |
| Italy | \(0.48\) | \(0.52\) | \(-0.03\) | \(0.52\) | \(-0.04\) | \(0.13\) | \(0.35\) |
Table A4 reports the per-country results. The gaps are positive nearly everywhere, and they are largest where fundamentals depart most from the U.S. baseline: 33 percentage points in Turkey, 23 in Brazil, 17 in Mexico, and 15 in Japan and Poland. Across most of Western Europe, whose prescribed shares sit close to the U.S. value, the gaps are within a few points of zero, with France exact to within a fifth of a point.
The hedged columns of Table A4 replace the local-currency volatility with the U.S.-dollar volatility of the index. With the premium common across countries, the hedged benchmark is 0.52 in every country. Mean advice lies within 6.2 percentage points of the hedged benchmark everywhere, with Brazil furthest below. The thirty-two-point spread of the unhedged benchmark, and with it every non-zero prescribed deviation \(\delta^{*}_{c}\) below, therefore comes from exchange-rate risk alone.
The home-index columns use each country’s own MSCI country index in local currency, with the premium and the volatility estimated over 2001 to 2025. The prescribed share then runs from 0.08 in Finland to 0.74 in Australia, a spread of sixty-six points, but the advice follows it only weakly. Over the eighteen non-U.S. countries with a series, the slope of the country effects on the home-index deviations is 0.11 (wild bootstrap 95 percent confidence interval \([0.04, 0.21]\)), against 0.14 for the world index on the same countries.
Section 5.5 reports the calibration slope, \(\hat{\kappa}=0.13\), and its leave-one-out range. The primary wild bootstrap over countries gives a 95 percent confidence interval of \([-0.145, 0.328]\) (standard error 0.112). The pre-registered influence diagnostics flag Brazil and Turkey, the two most volatile currencies, whose DFBETA values on the slope exceed \(2/\sqrt{N}\) in absolute value. The registered two-stage bootstrap, which resamples the thirteen models and re-estimates the twenty country effects on every draw, gives a standard error of 0.032 and an interval of \([0.067, 0.189]\). The small positive slope is therefore not first-stage noise, but with twenty countries it cannot be distinguished from zero. Both procedures place the slope far below one.
Measurement error in \(\delta^{*}_{c}\) would attenuate \(\hat{\kappa}\) toward zero, but sampling noise in the return moments is too small to matter. Resampling the twenty-seven return years (1999–2025) and recomputing every \(\delta^{*}_{c}\) on each draw moves the slope only from 0.13 to 0.10, although this addresses sampling error in \(\sigma_{c}\) and not the choices discussed in Section B.1.
The choice of risk aversion does not help match the advice and the benchmark. Because \(\delta^{*}_{c}\) is proportional to \(1/\bar{\gamma}\) wherever the benchmark is interior, as it is in every country on the grid, \(\hat{\kappa}\) is proportional to \(\bar{\gamma}\), rising from 0.09 at \(\bar{\gamma}=2\) to 0.43 at \(\bar{\gamma}=10\). Full calibration would require \(\bar{\gamma}\approx 23\), at which the benchmark prescribes no more than 7 percent equity in any country, while the level of the advice is consistent with \(\bar{\gamma}\) near 3.
B.5The mortgage benchmark
This subsection derives the benchmark of Section 4.2 for Question 2 (invest versus prepay a 4 percent mortgage). Prepaying a mortgage is a risk-free investment whose return is the after-tax mortgage rate. Because every return in this section is real, the prompt’s quoted nominal rate \(m\) must be deflated, so the return to prepayment is \[r^{m}_{c} \;=\; m\,(1-\tau^{m}_{c}) - \pi_{c},\] where \(\tau^{m}_{c}\) is the share of mortgage interest that is effectively deductible in country \(c\) and \(\pi_{c}\) is expected inflation, proxied by the trailing twenty-year mean of realized inflation. The expected equity return is \(\mu_{c}=r_{c}+(\mu-r)\), the local real safe rate plus the common premium.
Two country-specific channels operate here. First, deductibility differs sharply across the sample (broadly deductible in the United States for itemizers, no owner-occupier deduction in Germany). Second, expected inflation differs. A fixed nominal mortgage is a better real investment where inflation is low, as in Japan. Prepayment is the household’s safe asset only if it pays more than local government bonds, however. Where the after-tax real mortgage rate falls below the local real bond rate \(r_{c}\), the household should keep the mortgage and hold bonds for the safe part of its portfolio (Becker and Shabani 2010). The safe rate is therefore \(\max\{r^{m}_{c},\,r_{c}\}\), and the Merton logic gives \[\alpha^{Q2}_{c} \;=\; \min\Bigl\{1,\;\max\Bigl\{0,\;\frac{\mu_{c}-\max\{r^{m}_{c},\,r_{c}\}}{\gamma_{c}\,\sigma_{c}^{2}}\Bigr\}\Bigr\}, \tag{12}\] which is Equation (2) of Becker and Shabani (2010) with after-tax real rates.
The benchmark has two regions. Where \(r^{m}_{c}\) exceeds \(r_{c}\), prepayment raises the safe rate, so \(\alpha^{Q2}_{c}\) lies below the allocation benchmark \(\alpha^{S}_{c}\) of Equation (8), and full prepayment is optimal wherever the expected equity return \(\mu_{c}\) falls below \(r^{m}_{c}\). Where \(r_{c}\) exceeds \(r^{m}_{c}\), the mortgage is cheap debt that should not be prepaid, and \(\alpha^{Q2}_{c}=\alpha^{S}_{c}\). In this region the benchmark places the remainder of the savings in bonds rather than in prepayment. Note that this was not an option offered to the AI models, which had to either prepay or invest in stocks. At the baseline calibration the second region contains the five premise-acceptance countries, where the prompted 4 percent lies below local inflation, and Poland, where the two rates differ by less than half a percentage point. Because effective deductibility depends on income, itemization, and loan size rather than on a single statutory rate, I treat \(\tau^{m}_{c}\) as a calibration assumption rather than a point-sourced number and report \(\alpha^{Q2}_{c}\) across a \(\tau^{m}\in[0,0.5]\) band. The advice lies above the benchmark in every country even at the equity-favorable end of the band.
As mentioned above, the mortgage question offers only two uses of the savings: the index fund and prepayment. If the household could not hold bonds, the safe rate would be \(r^{m}_{c}\) alone. Table A5 reports this two-option benchmark \(\tilde{\alpha}^{Q2}_{c}\), Equation (12) with \(r^{m}_{c}\) in place of \(\max\{r^{m}_{c},\,r_{c}\}\), alongside \(\alpha^{Q2}_{c}\). The two coincide wherever prepayment pays more than bonds, so they differ only in the six countries above. Because \(\tilde{\alpha}^{Q2}_{c}\) is not bounded by \(\alpha^{S}_{c}\), it reaches 0.98 in South Africa, where the prompted rate is far below local inflation.
| Country | Mean advice | Benchmark \(\alpha^{Q2}_{c}\) | Band (\(\tau^{m}\in[0,0.5]\)) | Two-option \(\tilde{\alpha}^{Q2}_{c}\) | Gap \(\Delta^{Q2}_{c}\) |
|---|---|---|---|---|---|
| United States | \(0.83\) | \(0.47\) | \([0.41,\,0.52]\) | \(0.47\) | \(0.36\) |
| Japan | \(0.54\) | \(0.20\) | \([0.16,\,0.32]\) | \(0.20\) | \(0.34\) |
| France | \(0.67\) | \(0.33\) | \([0.33,\,0.52]\) | \(0.33\) | \(0.34\) |
| Sweden | \(0.73\) | \(0.39\) | \([0.28,\,0.46]\) | \(0.39\) | \(0.34\) |
| Switzerland | \(0.61\) | \(0.27\) | \([0.20,\,0.39]\) | \(0.27\) | \(0.34\) |
| Canada | \(0.70\) | \(0.37\) | \([0.37,\,0.52]\) | \(0.37\) | \(0.33\) |
| Germany | \(0.61\) | \(0.28\) | \([0.28,\,0.48]\) | \(0.28\) | \(0.33\) |
| United Kingdom | \(0.68\) | \(0.35\) | \([0.35,\,0.49]\) | \(0.35\) | \(0.33\) |
| Finland | \(0.63\) | \(0.32\) | \([0.32,\,0.51]\) | \(0.32\) | \(0.32\) |
| Poland | \(0.66\) | \(0.35\) | \([0.35,\,0.35]\) | \(0.38\) | \(0.31\) |
| Netherlands | \(0.67\) | \(0.44\) | \([0.31,\,0.50]\) | \(0.44\) | \(0.22\) |
| Spain | \(0.62\) | \(0.41\) | \([0.41,\,0.52]\) | \(0.41\) | \(0.21\) |
| Australia | \(0.65\) | \(0.45\) | \([0.45,\,0.49]\) | \(0.45\) | \(0.20\) |
| Italy | \(0.63\) | \(0.45\) | \([0.45,\,0.52]\) | \(0.45\) | \(0.18\) |
| South Korea | \(0.65\) | \(0.47\) | \([0.45,\,0.53]\) | \(0.47\) | \(0.18\) |
| China | \(0.60\) | \(0.47\) | \([0.45,\,0.48]\) | \(0.47\) | \(0.13\) |
| Turkey | \(0.81\) | \(0.21\) | \([0.21,\,0.21]\) | \(0.65\) | \(0.60\) |
| Brazil | \(0.81\) | \(0.23\) | \([0.23,\,0.23]\) | \(0.34\) | \(0.57\) |
| Mexico | \(0.81\) | \(0.38\) | \([0.38,\,0.38]\) | \(0.61\) | \(0.43\) |
| India | \(0.79\) | \(0.46\) | \([0.46,\,0.46]\) | \(0.79\) | \(0.32\) |
| South Africa | \(0.80\) | \(0.49\) | \([0.49,\,0.49]\) | \(0.98\) | \(0.31\) |
| \(\sigma_{c}\) | \(r_{c}\) | \(z_{c}\) | \(\rho_{c}\) | \(\bar{w}_{c}\) | \(R_{c}\) | \(L_{c}\) | \(g_{c}\) | \(\tau^{m}_{c}\) | \(\pi_{c}\) | \(\alpha^{S}_{c}\) | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Country | (%) | (%) | (std) | (%) | (000s) | (age) | (yrs) | (%) | (%) | (\(\bar{\gamma}{=}3\)) | |
| United States (USD) | 18.4 | 0.4 | 0.12 | 51.3 | 87 | 67 | 18 | 1.1 | 0.15 | 2.5 | 0.52 |
| United Kingdom (GBP) | 19.1 | -0.3 | 0.05 | 54.2 | 47 | 68 | 17 | 0.3 | 0.00 | 2.9 | 0.49 |
| Sweden (SEK) | 19.2 | -0.3 | 0.05 | 66.3 | 546 | 70 | 16 | 1.0 | 0.30 | 2.1 | 0.48 |
| Finland (EUR) | 18.5 | -0.1 | -0.28 | 65.7 | 51 | 68 | 17 | 0.5 | 0.00 | 2.0 | 0.52 |
| Germany (EUR) | 18.5 | -0.6 | -0.04 | 53.3 | 55 | 67 | 18 | 0.8 | 0.00 | 2.2 | 0.52 |
| France (EUR) | 18.5 | 0.2 | -0.03 | 70.0 | 44 | 65 | 22 | 0.6 | 0.00 | 1.9 | 0.52 |
| Spain (EUR) | 18.5 | 0.9 | -0.16 | 86.3 | 35 | 65 | 22 | 0.5 | 0.00 | 2.0 | 0.52 |
| Japan (JPY) | 20.4 | -0.1 | -0.36 | 42.4 | 5,186 | 65 | 22 | -0.2 | 0.10 | 0.9 | 0.43 |
| Brazil (BRL) | 27.5 | 0.8 | -0.25 | 97.5 | 56 | 64 | 20 | 1.2 | 0.00 | 5.5 | 0.23 |
| Australia (AUD) | 19.0 | 0.9 | 0.14 | 53.0 | 105 | 67 | 20 | 0.7 | 0.00 | 2.7 | 0.49 |
| Canada (CAD) | 18.5 | 0.3 | 0.18 | 45.1 | 86 | 65 | 21 | 0.9 | 0.00 | 2.2 | 0.52 |
| India (INR) | 19.6 | 1.0 | -0.28 | 44.6 | 273 | 58 | 23 | 4.9 | 0.10 | 6.4 | 0.46 |
| South Africa (ZAR) | 19.0 | 3.7 | 0.97 | 8.9 | 339 | 60 | 20 | 0.1 | 0.00 | 5.5 | 0.49 |
| Netherlands (EUR) | 18.5 | -0.5 | 0.19 | 96.0 | 61 | 70 | 15 | 0.1 | 0.35 | 2.4 | 0.52 |
| Italy (EUR) | 18.5 | 1.3 | -0.09 | 79.0 | 35 | 70 | 16 | -0.3 | 0.00 | 2.0 | 0.52 |
| Switzerland (CHF) | 18.7 | 0.2 | -0.02 | 47.5 | 98 | 65 | 22 | 0.7 | 0.20 | 0.5 | 0.51 |
| Poland (PLN) | 22.5 | 0.7 | -0.07 | 40.6 | 101 | 62 | 21 | 2.6 | 0.00 | 3.7 | 0.35 |
| South Korea (KRW) | 18.4 | 1.0 | -0.04 | 38.9 | 53,916 | 65 | 21 | 1.3 | 0.05 | 2.2 | 0.53 |
| China (CNY) | 19.1 | 1.4 | -0.02 | 103.6 | 106 | 60 | 22 | 6.8 | 0.05 | 2.2 | 0.48 |
| Mexico (MXN) | 21.7 | 2.9 | -0.14 | 79.6 | 277 | 65 | 18 | 0.1 | 0.00 | 4.4 | 0.38 |
| Turkey (TRY) | 29.1 | -4.2 | 0.02 | 96.4 | 743 | 64 | 18 | 5.7 | 0.00 | 19.4 | 0.21 |