4What Should Move the Advice?
Before turning to the main results, it is worth thinking about what should move the advice for investors in different countries. Holding the investor’s preferences, investment opportunities, and balance sheet fixed, the advice should not change merely because the investor names a different country or asks in a different language, unless there are cross-country differences that would warrant a different allocation. A lower recommended equity share in Brazil than in the United States would be inappropriate if the two investors face the same economic situation, but may reflect good advice if Brazil’s fundamentals warrant more conservative advice. This section asks which fundamentals the models should take into account. Later, I use the intuition developed in this section to separate three candidate advice rules: a Merton advisor that applies equation (1) to the stake alone, a Life-Cycle advisor that applies equation (2) to the household’s balance sheet, and a Questionnaire advisor that follows a retail risk questionnaire (Section 6).
To this end, I construct a simple benchmark to think through the economic fundamentals that would motivate country differences in advice. I also use a simple benchmark to evaluate the mortgage advice using the contract rate. Appendix B develops the benchmarks in full, and reports the country gaps under different hedging assumptions, the sensitivity of the results to risk aversion, and the influence diagnostics.
4.1A simple benchmark
The simplest benchmark treats the $50,000 stake as a standalone portfolio of a world equity index and domestic government bonds. For an investor in country \(c\) with constant relative risk aversion \(\gamma_c\), the optimal equity share is \[\theta^{*}_{c} \;=\; \frac{\mu_{c}-r_{c}}{\gamma_c\,\sigma_{c}^{2}}, \tag{1}\] where \(\mu_{c}-r_{c}\) is the expected excess return and \(\sigma_{c}\) the volatility of the equity fund, both measured in local currency. This is the rule of Merton (1969) and Samuelson (1969), where the share does not depend on the size of the stake or, with independent returns, on the investment horizon. If the uncovered interest parity holds, the equity premium is constant across currencies and countries. Under these assumptions, the optimal equity share across countries would vary only due to \(\sigma_{c}\) and \(\gamma_c\). Importantly, how much the benchmark varies depends on whether the world fund is hedged. For a currency-hedged fund and with a common risk aversion parameter across countries, the simple benchmark prescribes a share of 52 percent in equity everywhere. If we add exchange rate risk, the prescribed share then ranges from 0.21 in Turkey and 0.23 in Brazil to 0.53 in South Korea at \(\gamma=3\), a spread of thirty-two percentage points. I later return to whether actual advice tracks this spread.
This simple benchmark treats the stake as a standalone investment. If we add in more of the balance sheet, treat pension wealth and human capital as outside wealth \(B_{c}\), and decompose outside wealth into safe and risky assets, we can rewrite the optimal equity share as \[\alpha^{*}_{c} \;=\; \min\left\{1,\; \max\left\{0,\; \underbrace{\theta^{*}_{c}\left(1+\frac{B^{s}_{c}+B^{r}_{c}}{W}\right)}_{\text{equity demand on total wealth}} \;-\; \underbrace{\frac{B^{r}_{c}}{W}}_{\text{equity held outside}} \right\}\right\}, \tag{2}\] where \(\theta^{*}_{c}\) is the Merton share of equation (1), \(W\) is the stake, \(B^{s}_{c}\) is safe outside wealth, and \(B^{r}_{c}\) is outside wealth that carries equity risk. With safe outside wealth (\(B^{r}_{c}=0\)), which is an order of magnitude larger than the stake for an average 40-year-old, the prescribed share reaches the 100 percent corner in every country (Appendix B.2). With common risk aversion, the benchmark therefore varies across countries only through the currency risk of an unhedged fund.
Equation (2) organizes the tests in the rest of the paper around two properties. The first is invariance. Prompt content that leaves the economics unchanged should leave the advice unchanged. This covers the country name (Section 5.2), the language of the question and the local names of the instruments (Section 5.3), and, with independent returns, the investment horizon (Section 6). The only reason these variables should move the advice is if the models infer something about the investor or the problem. However, the models need to do that in a way that relates to the terms in the question.
This brings us to the second property, which is responsiveness. The terms inside the equation should move the advice. The volatility \(\sigma_{c}\) differs across countries for an unhedged fund. Risk aversion may differ across countries or for the individual investors. Note that risk aversion is not stated in the main prompts, but that I include it in the stated-investor arm. Safe outside wealth \(B^{s}_{c}\) rises with pension generosity and falls with age as human capital declines. Equity-like outside wealth \(B^{r}_{c}\) rises when the pension is invested in stocks.
The rest of the paper tests these two properties in turn. Section 5 tests invariance to the country name, the language of the question, and the local names of the instruments, and asks whether the country effects track fundamentals. Section 6 varies the investor’s age, risk tolerance, pension size and composition, and the investment horizon in the prompt. Table 2 in Section 6.4 collects the evidence on both properties.
Finally, there is a caveat to the invariance tests. The prompt does not say whether the fund is hedged, what taxes apply, or what else the investor owns, so a model may fill in these details from the country name. A country effect can therefore reflect a reasonable inference about local conditions instead of a failure of invariance. I return to this later.
4.2A benchmark for the mortgage question
The mortgage question involves different economics. Paying down a mortgage is a risk-free investment whose return is the after-tax real mortgage rate. In the question, the prompted rate is 4 percent, which should be adjusted for the deductible share of interest and expected inflation. Prepayment is a better investment only if this return exceeds the local real bond rate (Becker and Shabani 2010). Otherwise the household should keep the mortgage and hold bonds for the safe part of its portfolio. The relevant safe return is therefore the higher of the two rates. Replacing \(r_{c}\) in the Merton rule (1) with it gives the benchmark share \(\alpha^{Q2}_{c}\) (see Appendix B.5). Where the mortgage rate is the higher of the two, prepayment is a better investment when inflation is low and a worse one when interest is deductible. Where the bond rate is higher, \(\alpha^{Q2}_{c}\) equals the allocation benchmark \(\alpha^{S}_{c}\).
Because effective deductibility depends on income and itemization, the appendix also reports \(\alpha^{Q2}_{c}\) across a band of deductible shares from zero to one half. For the five premise-acceptance countries of Section 3, the after-tax real mortgage rate is negative and below the local real bond rate, and so prepaying is dominated by an investment in local bonds. This makes the mortgage benchmark equal to the simple benchmark above. I report these countries separately. The pre-registration does not specify this benchmark, so the comparison is exploratory. The benchmark takes the prompted 4 percent as given. Section 5.4 also reports advice when no rate is stated, for which no benchmark is defined.