Forex Automaton

AUD/JPY and GBP/USD form a weakly correlated pair, often but not always correlated positively, which many traders would regard to be one of the last suspects when it comes to predictive forex correlations. Nonetheless, in the screening we conducted last summer using the data set from 2002 to early 2008, AUD/JPY was found to form a leading indicator for GBP/USD. The feature was found by studying the time-integrated correlation of logarithmic returns in the two time series. As usual in such cases, a detailed time-evolution study is necessary to tell whether this effect is merely a result of a single high-impact event or a recurrent feature. I extend the period of observation up to April 2009, split it into three time windows of varying volatility, and analyze the time stability of the leading indicator effect.

This is a follow-up report on the predictive feature discovered in the correlation analysis of AUD/JPY and EUR/CHF. The feature was found by studying the time-integrated correlation of logarithmic returns in the two time series. A detailed time-evolution study is necessary to tell whether this effect is merely a result of a single high-impact event or a recurrent feature. The effect in question was found not to exist in the Asia-Pacific trading, therefore in this report I cover the trading hours from 4am to 8pm New York time, which excludes Asia. I extend the period of observation up to April 2009 and analyze the time stability of the leading indicator provided by AUD/JPY for EUR/CHF.

With USD/CAD, I am continuing the series of reports focusing in the time evolution of the forex correlation shapes during the present financial crisis. Extending the time coverage up to the end of March, I see the need to make the picture a bit more complex with three, rather than two phases with considerably different volatility level, since the volatility is seen to abate at the end of January 2009. The bipolar correlation pattern, a major subject of this research, is seen to disappear during the peak of volatility, but reappear later.

Pre-history dependence is what distinguishes forex from random walk. I analyze logarithmic returns in GBP/CHF on an hour scale and trace the time evolution of the auto-correlation peak throughout the present crisis, starting with August 2007. What I call "bipolar disorder" has emerged from this series of reports as a prominent pattern to accompany the high volatility phase of the crisis. Like in EUR/GBP, "bipolar disorder" in GBP/CHF had a sharp peak during the last two weeks of December 2008.

CHF/JPY dropped during the crisis, as did the interest rate differential between the currencies. The autocorrelation "pattern" of logarithmic returns resembles those of EUR/GBP, GBP/USD, EUR/JPY, and EUR/USD, with the hour-scale "bipolar disorder", seen in the analysis as a significant negative correlation value at one hour lag, growing in prominence during the volatile phase of the crisis. Still, in CHF/JPY it falls short of the magnitude seen in EUR/CHF (ironically, one of the least volatile in forex), not to mention AUD/USD. In short, not just the increased volatility discussed in the media, but also the "bipolar disorder", a particular form of pre-history dependence, is seen as the signature pattern of financial panic.

The "instanteneous" correlation between the ratio of S&P500/ Nikkei-225 stock market indexes and USD/JPY has been mostly positive during the crisis. As with GBP/USD vis-a-vis FTSE/DJIA, EUR/USD vis-a-vis DAX/DJIA, and EUR/USD vis-a-vis CAC/DJIA, one common feature is evident: one-day lag correlation "prefers" to have an opposite sign to the "instantaneous" correlation. For the case in hand, with the "instanteneous" correlation mostly positive, the stock index ratio seems to provide a stable leading indicator for USD/JPY suitable in principle to be incorporated into a trading algorithm.

From the two cases considered so far in the index-1/index-2 vs currency-1/currency-2 correlation technique (correlating FTSE/DJIA and DAX/DJIA with EUR/USD), one common feature is evident: one-day lag correlation "prefers" to have an opposite sign to the "instantaneous" correlation. Today's post is about correlations between CAC/DJIA (the ratio of France's and US' blue chip indexes, Cotation Assistee en Continu and the Dow) and EUR/USD.

I continue with the index-1/index-2 vs currency-1/currency-2 correlation methodology. Today's post is about correlations between DAX/DJIA (the ratio of Germany's and US' blue chip indexes, Deutscher Aktien IndeX and the Dow) and EUR/USD.

It just occured to me that a better way to look at a correlation between stock markets and forex (as always, in search for leading indicators) is to construct a forex-like ratio of two stock indexes and correlate that with the corresponding ratio of the nation's currencies. Therefore today I am studying correlations between FTSE-100/DJIA (the ratio of Britain's and US' blue chip indexes, Footsie and the Dow) and GBP/USD.

The correlation between AUD/JPY, the pure case of a carry-trade currency pair, and SPY, the S&P 500 ETF, measures the currency risk aversion effects on the US stock market -- and the position the stock market occupies on the agenda of forex carry-traders. The structure of the correlation peak (with time lag) on the day scale resembles the one seen in the hour scale studies of forex markets (same could be said of SPY alone). This makes one recall the topic of market fractality, or self-similarity, popularized by Mandelbrot ("The (Mis)behavior of Markets").

Having made so many references to bipolar disorder in the behavior of currencies, it seems appropriate, despite the forex focus of this research blog, to visit the market where the image of manic-depressive Mr Market was first conceived (by B.Graham): the stock market. I apply correlation analysis of logarithmic returns to the day data on the popular SPDR S&P 500 ETF, the SPY. This working paper on SPY autocorrelations begins a new series where currencies and other markets, such as interest rates, commodities, and stocks will be observed and studied together within a uniform framework of correlation analysis. Executive summary: 16 years of daily SPY data invalidate the behavioral pattern of waiting for and following Mr Market's "nod of approval" -- by showing the market, everything else being equal, to "self-contradict" and reward contrarians. This is particularly true in the extreme volatility environment of today.

As one more exchange rate analysis, the eighth one, is being added to the "patterns of crisis" series, the main conclusion holds. The extreme volatility of the past twenty or so weeks can not be fully grasped if characterized by variance alone, since such a characterization would omit important pre-history aspect in the variation of the forex quotes. The considerabe negative next-hour autocorrelation can be intuitively understood as the inclination of the market to contradict itself on the next-hour time scale. Therefore, the volatility is not just large, it is malignant, especially if viewed from the prospective of a human trader, as opposed to that of a non-anthropomorphic automated trading system algorithm: trend isn't your friend these days, at least on the time scale of hours.

As before, the time interval starting in August 2007 is split into two parts, or phases -- the "moderate" and the high volatility one, whose features are then compared. Among the currency pairs looked at so far in this series of "crisis" reports, EUR/CHF seems to be in a class of its own: unlike EUR/JPY, EUR/USD, and GBP/USD, and especially EUR/GBP, the autocorrelation patterns of the two phases have similar shapes. So they do for AUD/JPY, but in case of EUR/CHF, the "bipolar disorder" feature of the shape is pronounced more strongly. If this correlation shape is indeed a signature of the carry trade activity, one may conclude that the carry trade activity  in the less volatile EUR/CHF during the crisis remains popular.

Both the initial autocorrelation structure of EUR/GBP at the onset of the present crisis and its modification during the volatile phase of the crisis resemble those of EUR/JPY, EUR/USD, and GBP/USD: the autocorrelation function of logarithmic returns, with initially visible positive tails of the zero-lag peak, becomes oscillatory in the vicinity of zero lag as the EUR/GBP rally unfolds in the late 2008. Thus the abnormal volatility is accompanied by a specific correlation structure, and EUR/GBP is consistent with the broader picture of "bipolar disorder" correlation being the characteristic pattern of the financial panic.

In this series of posts I look for specific changes in the autocorrelation structure of leading forex exchange rates, related to the global financial crisis. The Holy Grail of automated trading system development is to find "stable imperfections" -- deviations from market efficiency which remain ever-present, or present at predictable times, or present most of the time without turning into their opposites the rest of time, or at lest present long enough for the AI to detect and "learn" them, use them, and "unlearn" them if they are no longer there without disastrous losses. From this point of view, you can share our interest in the "crises" of all sorts -- these are the periods when some of the rules that worked in the past no longer do. A burning question is whether the episodes of panic are characterized by more order or less order, compared to "normal" times, and whether this is the "same" order or the rules of order are changed. The previous case studies, dealing with EUR/JPY, EUR/USD, GBP/USD, had a common conclusion: the rules of crisis are different, and seem to advance a particular correlation pattern, which I called "bipolar disorder". The case of GBP/JPY was not so clear, possibly because GBP/JPY was seen to be possessed by "bipolar disorder" even in a late 2002 -- early 2008 analysis. In the case of AUD/JPY, the "bipolar disorder" feature was seen to be most pronounced in 2004 (cheap money period) and to recede already in 2007, which made me associate it tentatively with carry trade. For these high interest differential exchange rates, the "bipolar disorder" seems to be the hallmark of "good times", times of high risk tolerance. This complicates the interpretation. To make things even more complex, for the Aussie (as for the Sterling) the crisis brought about a considerable reduction of the interest rates, thus changing the playground as far as the rate differential games are concerned. Nevertheless, below are the details for AUD/JPY.

The picture that emerges from the autocorrelation analysis of the GBP/JPY time series on the hour scale looks consistent with what has been previously seen with EUR/JPY, EUR/USD, and GBP/USD. To say that these exchange rates became dramatically more volatile in the second half of 2008 would be to say too little, since one could hypothetically observe very high volatility in a martingale manner, that is without any prehistory dependence whatsoever. The actual pattern of crisis, on the contrary, seems to be characterized by a prehistory dependence of its own, whose essence is the reduced, compared to a hypothetic fully unpredictable time series (no matter how volatile!), likelihood of sustained trends. This is the environment where a trend reversal bet is, everything else being equal, more likely to succeed than a trend continuation bet. To repeat, I am talking about GBP/JPY autocorrelation structure in the hour scale analysis context, not whether the longer-scale movements of the second half of 2008 could be predicted by technical analysis.

The GBP/USD study is done in the same way as the previous EUR/JPY and EUR/USD studies: the time interval starting in August 2007 is split into "moderate" and high volatility parts, whose features are then compared. The rally of low-yielders (USD and JPY) that characterized the high volatility phase seems to distort the autocorrelation structure in a very particular way, and this influence is found to be qualitatively the same for the exchange rates looked at so far.

I continue the "topical" theme of financial crisis, namely how it's seen through the optics of correlation analysis. With the data so far it looks like the crisis does have a unique correlation aspect to it: the increased volatility of the "impact phase" is not just large, but seems to be accompanied by a particular (non-random) correlation structure. Rather than being a "random walk" with a long step, it's a walk where each step (abnormally large indeed) is slightly more likely to be in the direction opposite to that of the previous step. As before, I use logarithmic returns on the hour time scale.

This article is first in a series dedicated to the impact of the present financial crisis on the correlation patterns of the popular currency exchange rates. The acute phase of what began as the subprime mortgage crisis in 2007 has, among other things, challenged the ability of financial models (of the variety practiced by large institutions some of which no longer exist) to be of value to their users. Leaving CDOs and credit default swaps aside, with the forex and interest rate data at hand here on the Forex Automaton site, and with the same set of tools as before, I am going to approach the following questions. What exactly changed during the crisis? Did the markets become unpredictable? Less predictable than usual? More predictable in a new way? Was it a qualitative change in the set of predictability patterns? Or was it only a change in the amplitude of the fluctuations? I begin with EUR/JPY, an exchange rate which recently suffered some of the most dramatic changes seen in forex.

This is a follow-up report on the predictive feature discovered in the correlation of AUD/JPY and EUR/USD. To those who are new to this site: the blog focuses on applying quantitative analysis to forex in search for significant patterns which can be utilized for trading short range, the minimal time scale being one hour.

The feature in question was found not to exist in the Asia-Pacific trading, therefore in this report I cover the trading hours from 4am to 8pm New York time, which excludes Asia. I extend the period of observation up to the fourth quarter of 2008 and analyze the time stability of the delayed correlations between AUD/JPY and EUR/USD.