Effectiveness of foreign exchange interventions: evidence from New Zealand

ABSTRACT

This paper examines the effectiveness of explicit and implicit foreign exchange (FX) interventions in New Zealand: one secret spot market intervention and two implicit interventions – a regular Monetary Policy Statement (MPS) and an unexpected oral intervention by the Reserve Bank of New Zealand (RBNZ) governor addressing the New Zealand Dollar (NZD). By applying a synthetic control methodology to a unique dataset of RBNZ interventions, we construct a counterfactual to estimate their effect. The results indicate that the actual intervention and the MPS release were ineffective in moving the NZD. However, the speech depreciated the NZD by 1.12%, although the effect was small and short lived. Our findings suggest that FX interventions, explicit or implicit, are a weak policy tool to affect the exchange rate in New Zealand.

KEYWORDS:

• Foreign exchange intervention
• explicit intervention
• implicit intervention
• synthetic control
• effectiveness

JELS Codes:

• C31
• E58
• F31

Acknowledgements

We would like to thank the Reserve Bank of New Zealand for allowing us access to the dataset that made this paper possible. We also would like to thank Anella Munro, Ken Li, Severin Bernhard, the seminar participants at the Reserve Bank of New Zealand, and the University of Canterbury for valuable insights and feedback on this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 However, in the past, the Bank of Japan has followed the approach of ‘leaning with the wind’ and a similar perspective has also been contemplated by the Reserve Bank of New Zealand (De Gregorio & Tokman, 2005).

2 Other pragmatic reasons for intervention include building up foreign reserves and maintaining market liquidity or even profitability (Cassino & Lewis, 2012; Neely, 2001).

3 Data on the RBNZ’s interventions are not publicly available, leaving an absence of research on this topic. There are, however, three related papers. Karagedikli and Siklos (2008) investigated how monetary policy, especially surprise announcements, affected the NZD, finding that monetary policy surprises typically had a permanent impact on the exchange rate. The second study, by Cassino and Lewis (2012), considered the profitability of the RBNZ’s FX interventions, discovering that they can be profitable, although most of this profit is driven by building up FX reserves. The final relevant New Zealand paper examined the effect of statements from the RBNZ on the interest rate (Guthrie & Wright, 2000). It concluded these statements had led to significant changes in New Zealand interest rates that could not be attributed to open market operations.

4 An intervention might occur as a separate event or as an intervention cluster when the RBNZ intervened on consecutive days.

5 Note that we cannot disclose the exact date of the first intervention when the NZD was sold by the RB.

6 There are three events worth commenting on during this period. On 4 September, the European Central Bank unexpectedly cut rates by 10 basis points (European Central Bank, 2014; Udland, 2014), the Bank of England held the bank rate at 0.5% (Bank of England, 2014), and the Bank of Japan (Bank of Japan, 2014) held rates. An election was held in New Zealand on 20 September. Finally, the RBNZ released the foreign currency assets and liabilities on 29 September. The net NZD sales would have indicated that the RBNZ had intervened in August, although the market would not have known the exact date. We examined the NZD around each of these other events and found little reaction.

7 The Hong Kong dollar was eliminated since it operates a currency board (International Monetary Fund, 2018).

8 A placebo that deviates before the intervention cannot be a result of it. These placebos add no value to the inference of success and are removed. A systematic rule is used to remove the ‘bad’ placebos. The MSPE for each placebo is calculated for the three days before the intervention. A ‘good’ placebo will have a small MSPE for this period. If the MSPE from a placebo is larger than the 99th percentile of the MSPEs produced from all the placebos, it is dropped.

9 Two robustness checks were used to check the consistency of these results. The 20-day control period was lengthened to 30 days to verify that the shorter control was sufficient, and the speech remained the only significant event. An AR(1) model was used as an alternative to forecast a counterfactual during each of the three events. Just as with the synthetic control method, the effect of the speech was the most apparent.