# RDP 2009-09: Volatility in International Capital Movements Appendix A: Predictability of Capital Flows

December 2009

In this Appendix we employ some simple econometric techniques to check the robustness of the findings discussed in the main part of the paper.

## A.1 Forecasting

One way of ascertaining whether knowledge of a particular flow conveys information useful in making inferences about the overall capital account is to test how well it can explain contemporaneous capital account developments. In particular, we test whether the fit of a naïve first-order autoregressive (AR(1)) model of the capital account can be improved upon by including information about specific components of the capital account by estimating the equation

where *kabr _{t}* is the capital account in ratio to GDP in the current
period, is the contemporaneous value of the

*i*th capital flow as a ratio to GDP, and

*ε*is the error term. We run this regression for the baseline AR(1) model, and for each of the five key components of the capital account.

_{t}
Table A1 reports the main results of these regressions for industrialised and emerging
economies. The Root Mean Squared Error (RMSE) serves as a measure of the fit
for each model. Results are reported in terms of the ratio of the RMSE of the
*i*th model as a ratio to the RMSE of the naïve baseline AR(1)
model. A value of 1 signifies no improvement over the naïve model. A value
greater than 1 signals deterioration in the ability to predict the capital
account, while a value less than 1 implies an improvement (adjusted for degrees
of freedom). In the interest of brevity we do not report the ratio for every
flow and every country. Instead we distinguish between industrialised and emerging
economies and average the ratios across the six countries in each sample. The
second column lists the countries for which we find a statistically significant
coefficient on the variable of interest.

Model |
RMSE_{i}/RMSE_{Naïve} |
Countries for which p-value indicates significance at 5 per cent level |
---|---|---|

Industrialised economies | ||

Foreign direct | 0.999 | – |

Portfolio equity | 0.991 | Japan, UK |

Portfolio debt | 0.991 | Japan |

Banks & money markets | 0.992 | Germany, Sweden |

Reserves | 0.994 | Australia |

Emerging economies | ||

Foreign direct | 1.001 | – |

Portfolio equity | 1.004 | – |

Portfolio debt | 0.990 | Mexico |

Banks & money markets | 0.922 | Korea, Thailand, Mexico, Argentina |

Reserves | 0.990 | Brazil, Mexico |

Notes: Simple average of results for industrialised and emerging economies. Due to data availability, regressions for Korea, Mexico, Argentina and the Philippines are calculated on samples starting in 1988:Q1, 1989:Q1, 1992:Q1 and 1996:Q1. All samples end in 2005:Q4. |

As expected, all industrialised and emerging economies have highly significant coefficients on the lag of the capital account in the naïve model. However, for emerging economies the average RMSE for the naïve model is more than twice as large as that for industrialised countries (not shown). Introducing information on a particular type of flow generally adds less than 2 percentage points of explanatory power to the model (adjusted for degrees of freedom) for industrialised economies, and less than 2½ percentage points for emerging economies (excluding the bank and money market flows), and in many cases there is an outright deterioration in explanatory power.

For industrialised economies, individual flows do not consistently add significant explanatory power over and above the naïve AR(1) benchmark. Portfolio equity flows are statistically significant for Japan and the United Kingdom, but not for other industrialised economies. Portfolio debt flows are only significant for Japan, and bank and money market flows only for Germany and Sweden, and in all of these cases the fit is only marginally improved. Information on reserve flows markedly improves the fit for Australia, but not for any other industrialised economies.

Emerging economies on the other hand appear to exhibit a more robust relationship between bank and money market flows and the capital account, perhaps reflecting a degree of bank dependence, as discussed earlier. Bank and money market flows improve the fit over the naïve model for Korea, Thailand, Mexico and Argentina. However, for the Philippines and Brazil the coefficient on bank and money market flows is not statistically significant. Other types of capital also fail to consistently improve the fit of the model. In summary, these results suggest that adding information about any particular flow is not particularly useful when trying to understand and predict capital account developments.

## A.2 Offsetting versus Compounding Relationships

An additional way to gauge the substitutability of particular flows within the capital account is to examine whether a flow tends to move in the same direction as the rest of the capital account – compounding the movement of the other flows – or if it exhibits an offsetting relationship. To capture these relationships, we regress each flow on the sum of the remaining flows within the capital account as follows

where: is the local currency value of the *i*th capital flow as a ratio to GDP in
the current period; is the *j*th capital flow as a ratio
to GDP in the current period; and *ε _{t}* is the error
term.

A negative coefficient implies an offsetting relationship, while a positive coefficient indicates that the flow has a compounding effect on the rest of the capital account. For each of the industrialised economies, almost all flows are offsetting to some degree, with virtually all coefficients negative and statistically significant (Table A2). Bank and money market flows tend to have the strongest negative relationship with the rest of the capital account in industrialised economies. For the United States, reserve flows do have a positive and statistically significant relationship with the rest of the capital account. However, this relationship is economically insignificant compared to all other relationships reported in this section.

Dependent variables | US | Japan | Germany | UK | Australia | Sweden |
---|---|---|---|---|---|---|

Foreign direct investment | −0.15^{**} |
−0.17^{**} |
−0.45^{**} |
−0.84^{**} |
−0.74^{**} |
−0.66^{**} |

Portfolio equity investment | −0.07^{*} |
−0.73^{**} |
−0.51^{**} |
−0.83^{**} |
−0.71^{**} |
−0.63^{**} |

Portfolio debt investment | −0.31^{**} |
−0.86^{**} |
−0.43^{**} |
−0.94^{**} |
−0.84^{**} |
−0.49^{**} |

Bank and money markets | −0.19^{**} |
−0.92^{**} |
−0.86^{**} |
−0.87^{**} |
−0.83^{**} |
−0.80^{**} |

Reserves | 0.01^{*} |
−0.67^{**} |
−0.15^{**} |
−0.14^{**} |
−0.80^{**} |
−0.28^{**} |

Notes: ** and * denote significance at the 5 and 10 per cent levels respectively; quarterly observations. All samples end in 2005:Q4. |

For emerging economies, almost all coefficients are also negative and statically significant. However, the offsetting relationship between particular flows and the rest of the capital account is generally less pronounced, particularly for direct investment and portfolio equity flows (Table A3). The exception is reserve flows which, on average, have a stronger offsetting relationship in the emerging economies examined than in the industries economies. Similar to the results for industrialised countries, bank and money market flows tend to have a strong negative relationship with the rest of the capital account.

Dependent variables | South Korea | Thailand | Philippines | Brazil | Mexico | Argentina |
---|---|---|---|---|---|---|

Foreign direct investment | −0.10^{**} |
−0.15^{**} |
−0.15^{**} |
−0.16^{**} |
−0.17^{**} |
−0.24^{**} |

Portfolio equity investment | −0.40^{**} |
−0.04^{**} |
−0.10^{**} |
−0.14^{**} |
0.01 | −0.19^{**} |

Portfolio debt investment | −0.16^{**} |
0.07^{**} |
−0.37^{**} |
−0.59^{**} |
−0.54^{**} |
−0.38^{**} |

Bank and money markets | −0.55^{**} |
−1.07^{**} |
−0.56^{**} |
−0.71^{**} |
−0.42^{**} |
−0.69^{**} |

Reserves | −0.61^{**} |
−0.38^{**} |
−0.39^{**} |
−0.80^{**} |
−0.65^{**} |
−0.61^{**} |

Notes: ** and * denote significance at the 5 and 10 per cent levels respectively; quarterly observations. Due to data availability, regressions for Korea, Mexico, Argentina and the Philippines are calculated on samples starting in 1988:Q1, 1989:Q1, 1992:Q1 and 1996:Q1. All samples end in 2005:Q4. |

As emerging economies typically have more volatile capital accounts than industrialised economies (see Section 2.1), the results of this section may indicate that the degree to which individual flows have offsetting relationships with the rest of the capital account is an important factor in determining overall volatility. In addition, our results suggest that bank and money market flows may play an important role in reducing capital account volatility as they tend to offset the combined developments in the remaining flows to a greater degree than any other type of flow.

The relatively strong offsetting relationship between bank and money market flows and the rest of the capital account may reflect the fact that banks perform an important intermediation function in both industrialised and emerging economies. They are typically also active in international debt and foreign exchange markets, not only as intermediaries, but also as an important source of arbitrage. Perhaps it is this function which allows bank and money market flows to readily adjust and offset developments in other flows.