I have been reading Michael Howell’s *Capital Wars*. Howell is a former Solomon Brothers strategist and founder of CrossBorder Capital. The book offers a very sophisticated take on global finance — definitely not for beginners — and provides a counterweight to Matt Klein and Michael Pettis’ *Trade Wars are Class Wars*. The latter, in my opinion, analytically overweights net saving flows and doesn’t pay enough attention to gross funding flows. You can see my long exchange on this issue with Matt on Twitter:

There is much I am persuaded by in *Capital Wars*; some I am not. But I am still reading it, so I’ll hold my judgement. What caught my eye was the predictive information contained in their forex risk index, which is essentially constructed by subtracting private sector liquidity from central bank liquidity. The measure predicts one-year ahead trade-weighted exchange rate of the dollar:

It is not exactly clear how they are constructing the index. But this kind of predictive ability is unheard-of in finance. Predicting exchange rate movements is especially hard. Part of the problem with this surprising scatter plot in the book is surely that they haven’t detrended the trade-weighted dollar index. That’s not kosher. You should at least first-difference the series to remove the autocorrelation. (The same problem plagues Turchin’s empirical evaluation of the Structural-Demographic Theory.)

I have previously argued that cross-border banking flows provide a good barometer of the global financial cycle. In the upswing of the global financial cycle, cross-border funding by global banks expands rapidly; in the downswing, it slows down or shrinks outright. That is why we expect it to track the tidal whipsaw of the global financial cycle. The surprising scatter plot in the book made me curious about the predictive information about future dollar strength contained in cross-border banking flows. We would like to test this hypothesis. We know that a stronger dollar corresponds to tighter financial conditions globally; conversely, a weaker dollar signals greater ease of finance. So the precise hypothesis we want to test is whether greater global liquidity, as captured by positive shocks to cross-border banking flows, predicts a weaker dollar. And we want to test this against the null hypothesis that, as standard macroeconomics would have it, the US trade balance governs the future strength of the dollar.

We obtain the (narrow, 27 economies) trade-weighted dollar index from the Bank of International Settlements. Also from the BIS, we obtain global liquidity, or liquidity, defined as the global aggregate of ‘cross-border claims denominated in all currencies plus local claims denominated in foreign currencies’. We obtain the US current account balance (“balance”) from Haver Analytics. In order to detrend the three series, we log transform liquidity and first-difference all three. All data is at the quarterly frequency. Figure 1 displays our three variables of interest before any detrending.

In order to test our hypotheses, we fit a vector autoregression model and perform Granger causality tests. Put simply, what we are doing is regressing or projecting each of the three variables onto lagged values of all three. The null of the Granger test is that a time-series does not Granger-cause the other, that is, it does not predict the other, after controlling for its own lagged values. If we can reject the null with high confidence, we have solid evidence that one series predicts another. Before we get started, we present the detrended series.

Using AIC or BIC, we can see that no more than 1-qtr lags are necessary to get obtain a good fit. Table 1 presents our parameter estimates.

Table 1. Vector autoregression parameter estimates. | |||

Results for equation Dollar | |||

Coef | StdErr | P | |

L1.Dollar | -0.02 | 0.07 | 0.75 |

L1.Liquidity | -37.77 | 6.68 | 0.00 |

L1.Balance | -0.01 | 0.02 | 0.55 |

Results for equation Liquidity | |||

Coef | StdErr | P | |

L1.Dollar | 0.00 | 0.00 | 0.92 |

L1.Liquidity | 0.22 | 0.08 | 0.01 |

L1.Balance | 0.00 | 0.00 | 0.67 |

Results for equation Balance | |||

Coef | StdErr | P | |

L1.Dollar | 0.03 | 0.38 | 0.94 |

L1.Liquidity | -17.81 | 33.88 | 0.60 |

L1.Balance | 0.01 | 0.08 | 0.90 |

Source: BIS, Haver Analytics, author’s computations. Estimates in bold are significant at the 1 percent level. |

We can see right off the bat that only liquidity predicts future changes in the value of dollar. In particular, the current account balance does not. We can also see that only liquidity is autocorrelated. Now we formally test this result with Granger causality tests. Table 2 presents our results. We have potential predictors in the rows, and variables that they hope to predict in the columns. The first panel displays the F-statistics for the Granger causality tests; the second displays the p-values for the same.

Table 2. Granger Causality Tests. | |||

F-statistics | |||

Dollar | Liquidity | Balance | |

Dollar | 0.01 | 0.01 | |

Liquidity | 31.96 | 0.28 | |

Balance | 0.35 | 0.18 | |

P-values | |||

Dollar | Liquidity | Balance | |

Dollar | 0.92 | 0.94 | |

Liquidity | 0.00 | 0.60 | |

Balance | 0.55 | 0.67 | |

Source: BIS, Haver Analytics, author’s computations. Estimates in bold are significant at the 1 percent level. |

We find that the current account balance does not predict the future value of the dollar but our measure of liquidity does. How strong is the result? The next figure displays the predictions against the observed values of changes in the dollar index. This level of predictability is almost unheard-of in finance. Clearly, our measure of global liquidity, the flux of international banking flows, contains a very strong signal of the global financial cycle.

We also display time-series plots for the observed and 1-qtr ahead predicted values. The observed values of the detrended dollar index appear to be much more volatile than the predicted values. This is as it should be since we can only predict a sixth of the variation in the series — *which is a lot!*

The results presented here are related to our earlier report on the structure of global markets. If cross-border banking flows contain a strong signal of the global financial cycle, then national stock markets, whose structural relationships we examined there, should be sensitive to the strength of the dollar and shocks to our measure of global liquidity. When the dollar goes up, they should go down; and conversely. Similarly, positive shocks to global liquidity should be associated with positive stock market returns.

This is indeed the case. Positive shocks to the dollar are negatively associated with national stock market returns. The only exceptions are three center countries (Holland, Switzerland and Sweden) where the association is not statistically significant. Note also that more peripheral countries are more exposed to dollar strength.

We also find that positive shocks to global liquidity are positively associated with national stock market returns. And we find again that more peripheral markets are more exposed to global liquidity shocks. The whipsaw of the global financial cycle is at its most brutal in frontier markets.

We have been working at the quarterly frequency because that’s the frequency at which BIS data is available. But there is no reason to believe that the data generating process itself — the global financial cycle — lives at that frequency. Indeed, by working at the quarterly frequency, we have averaged away most of the variation that exists. It seems likely that the global financial cycle governs day-to-day or even intra-day fluctuations in the strength of the dollar and global asset values. Can we find a proxy for cross-border banking flows at higher frequencies? That would be most useful.