TRY (Turkish Lira) Intermarket Correlation: USD/TRY Follows EUR/JPY

A time-integrated study of hourly time-scale lagged correlation between USD/TRY and EUR/JPY reveals hints of a correlation whereby EUR/JPY leads and TRY/USD follows with a lag on one hour.

Compared to the series of intermarket correlation reports generated in 2008-2009 (not covering Turkish Lira), the new reports adopt a somewhat different approach to statistical uncertainty estimation. Instead of synthesizing mock time series with the volatility distribution of the real ones, producing their autocorrelations (and intermarket correlations), and inferring the uncertainty from a comparison of many such analyses, as was done in 2008-2009, I now infer the uncertainty of the correlation coefficients directly as a standard deviation of the sum, knowing the terms entering the sum. As before, hourly data are used and the quantity correlated is hourly logarithmic return.

The data used cover the period from November 14, 2010 till December 09, 2012.

USD/TRY and EUR/JPY hourly correlation 1.1 USD/TRY and EUR/JPY hourly correlation zoomed 1.2

Fig.1. Correlation in hourly logarithmic returns between USD/TRY and EUR/JPY, 2010-2012. 1.1: Lags from 0 to 200 hours. 1.2: Zooming on the signal. The correlation is normalized so that total correlation corresponds to correlation strength of 1, total anti-correlation — to correlation strength of -1 (see Pearson correlation coefficient).

The time lag between the USD/TRY and EUR/JPY time series is defined as

td = tUSD/TRY – tEUR/JPY.

The negative correlation content in the +1 hour time lag bin is seen with about 2.5 standard deviations. This means that a move (the present analysis is insensitive to the direction of that move) in USD/TRY and time t and a move in EUR/JPY at time t-1 are negatively correlated over the time of observation. This is the same as saying the TRY/USD follows (trails) a movement in EUR/JPY with a lag of one hour. Due to the discrete nature of binning, the actual ticks that create the effect could be separate by a time interval from anything about zero to anything below two hours and still land in the separate adjancent hourly periods, creating a 1-hour lag effect. The average such time interval is one hour.

USD/TRY and EUR/JPY hourly autocorrelations, zoomed

Fig.2. Autocorrelation in hourly logarithmic returns in USD/TRY and EUR/JPY, 2010-2012. The correlation is normalized as in Fig.1.

Speaking of pair trading, when forming a pair of EUR/JPY and USD/TRY, due to the negative correlation, both positions must be held short or long at the same time. The relative weight with which EUR/JPY and USD/TRY enter the pair must be optimized to increase the one-lag correlation with respect to the one at lag zero. Qualitatively speaking, the features around zero in Fig.2 (zero autocorrelation at 1-hour lag in EUR/JPY and negative autocorrelation at 1-hour lag in USD/TRY) will not cancel the feature in Fig.1 (also negative). The resulting pair will be a mean-reverting autocorrelated time series.

Central European Time

Central European time is chosen for the following reason. Forex week begins, roughly speaking (since the volume increase is gradual) on Sunday 5pm and ends Friday 6pm Eastern time. It is convenient to define this week to consist of 5 full days, from 6pm Sunday to 6pm Friday New York time. When it’s 6pm in New York, it’s midnight in Berlin, Paris, Madrid, Rome, Geneva and Frankfurt. These cities use Central European Time or CET. Therefore, the convenience of using CET is that one gets 5 non-interrupted, full 24-hour long trading days per week. Table 1 compares four time zones including major trading centers of the world.

Tokyo91011 121314 15161718192021 2223012 3 456 78
Central Europe12345678910111213141516171819202122230
Greenwich01234567891011121314151617181920212223
Eastern US19202122230123456789101112131415161718

Table 1. Time zone conversion table. Seasonal time shifts, such as daylight saving time, may complicate the picture if the nations choose to enact them on different days, and are ignored.

Profile Histogram

A profile histogram provides an economic representation of two-dimensional (X vs Y) data. Unlike a two-dimensional histogram, the profile histogram is unbounded in the Y (vertical) dimension. Like any histogram, it has discrete bins along the X axis to aggregate the data. The data are represented graphically as a series of (X,Y) points on a plot, with the point position in Y corresponding to the mean and the vertical bars typically indicating the measure of the spread such as the RMS, or the measure of the precision of the mean, whereby larger bars correspond to less precision.

None of the projection techniques is perfect, since the reduction of information involved in all projections is not guaranteed to be “intelligent”. But they do solve the problem, even though it is necessary to look from more than one point of view and try more than one path to optimization.

High Order Cumulants

Cumulants are statistical measures of correlation designed to go to zero whenever any one or more quantities under study become statistically independent of the rest. Cumulants generalize the concept of a correlation measure; in particular, a correlation of two bodies, quantities and so on, the most intuitive one, can be represented and measured by the second-order cumulant. Higher orders can be conceived.

Financial time series come as sequences of bars or “candles”, one bar per time step of the series. The bar has an open, close, low and high levels of price. In the liquid market like forex, close, low and high should be sufficient while open is typically not too different from the previous close and is believed to be redundant.

To judge the quality of market predictions, we are interested in multivariate cumulants, since for each of the three essential components of a candle there is a prediction. Because we make predictions for each of these components, the number of variables we would like to correlate is even, and therefore we are interested in even-order cumulants.

The simplest of these is second order cumulant also known as covariance:

c2 = E[x1x2] – E[x1]E[x2] (1)

Here E[] stands for the averaging (a.k.a. expectation) operator.

In general, n-th order cumulant is constructed by making a correlation term E[x1x2…xn] and subtracting all terms which are not “genuinely” n-th order correlations, but are composed of lower order ingredients.

Take for example the daily candle. We can generate predictions for daily changes in low and high, such that the correlation between the real and predicted change for high will be positive. Same for low. If we take a trading position having yesterday’s low as a stop-loss and yesterday’s high as a profit target, we want to make sure that not only there is a tendency for low not to be hit when the prediction says so (attested to by the positive correlation between the daily change in low and its forecast), and not only there is a tendency for the high to be hit when the prediction says so (attested to by the positive correlation between the daily change in high and its forecast), but that these two things tend to happen “simultaneously” within the same trade. This is the essence of the difference between the “genuine” fourth order correlation and a mere superposition of two second order ones.

Forth order cumulant is defined as:

c4 = E[1,2,3,4]
– E[1,2,3]E[4] – E[1]E[2,3,4] – E[1,3,4]E[2] – E[1,2,4]E[3]
– E[1,2]E[3,4] – E[1,3]E[2,4] – E[1,4]E[2,3]
+ 2(E[1,2]E[3]E[4] + E[1,3]E[2]E[4] + E[1,4]E[2]E[3] + E[2,3]E[1]E[4] + E[2,4]E[1]E[3] + E[3,4]E[1]E[2])
– 6E[1]E[2]E[3]E[4].
(2)

Here we use the notation: E[x1x2…] gets replaced by E[1,2…] for the sake of brevity.

What is being subtracted is in fact products of lower order cumulants, which in turn subtract their lower order cumulants, which is why there are terms with both plus and minus sign alternating in a certain order. A recurrence relation exists allowing one to express higher order cumulants in terms of lower order ones.

A cumulant of order higher than 2 will go to zero if any two quantities are proportional to each other:

x1=ax2. (3)

The fact that it will also go to zero whenever any one quantity is statistically independent of the rest, combined with the additivity of cumulants, implies that a higher order cumulant will go to zero whenever any pair of quantities has even a less deterministic, randomized form of that equation:

x1=ax2 + r (4)

where r is a random number independent of x2.

A non-zero higher order cumulant indicates that a relationship between the data is not merely Eq. (4), with its familiar visualization as a diagonally elongated cloud in the x1, x2 space — even though one may see such a cloud and other signatures of two-point correlations when subjecting higher-order correlated data to a lower order analysis.

To keep the cumulant independent of the units in which the underlying quantities are expressed, we sometimes normalize it:

C4 = c4/(Var[1]Var[2]Var[3]Var[4])1/2, (5)

where Var is variance.

Forex Automaton as a Shannon’s Communication Channel. Introducing Kelly Criterion.

The intention of this post is to tie together several topics which appeared on my radar screen in the course of the trading system optimization. First, it has been understandably hard to fully rid oneself of vestiges of the mainstream financial theory based on the postulate of market efficiency, while building a wealth-generating tool relying explicitly on demonstrable market inefficiencies. The realization that Sharpe ratio does not let one make an objective choice of a portfolio was there from the beginning, and I recall perceiving this fact as a “necessary evil”. Then came the understanding of the fact that an arithmetic average of returns gives one a biased picture of long-term return, and consequently, Sharpe ratio is built around biased quantities.

In what followed, the key concept was that of Kelly Criterion and the rich intellectual context it is part of — quite remote from the mainstream financial engineering. Initially I came across a mention of Kelly Criterion in some pieces by Edward Thorp, found on his website, but did not fully appreciate the depth of the context. Later, prompted by some readers of this site, I learned about Ed Thorp’s more extensive exposition of Kelly Criterion, “The Kelly Criterion in blackjack, sports betting, and the stock market”.

A good place to begin is probably Claude Shannon’s work The Mathematical Theory of Communication (originally, The Bell System Technical Journal, Vol. 27, pp. 379-423, 623-656, July, October, 1948). The work introduces the concept of information in the strict theoretical sense. It deals with measures of information and redundancy. These are probably two most important concepts in the algorithmic trading — suffice it to say that if the markets had not been redundant, algorithmic trading in the sense discussed and developed here would have been impossible and trading would degenerate into gambling. Much of ForexAutomaton.com’s research content directly deals with measurements of informational redundancy in forex. Shannon’s information theory is, among other important things, the venue for cryptography, linguistics, statistics and mathematics to meet in the context of financial speculation. Shannon introduced the concept of mutual information to characterize transmission capacity of communication channels.

In 1956, Kelly introduced what is known as Kelly Criterion in an application of Shannon’s theory to gambling. He titled the work “A New Interpretation of Information Rate”. The communication channel considered is a very specific one: it is a noisy channel allowing a gambler to know in advance the outcomes of chance events and bet accordingly. “Noisy” means that the information gambler receives is in general not perfect. Shannon’s mutual information (channel capacity) is the measure of quality of betting tips the gambler receives. Because the chance of losing is non-negligible, the gambler can not bet all of her capital on every game, but because the value of insider data she receives is non-negligible either, her optimal betting ratio is non-zero. The optimal betting ratio in the simple case of two symmetric outcomes equals the difference between the win and loss probabilities if the tip is followed. The maximum expected logarithmic rate of growth of capital (per game) turns out to be Shannon’s mutual information of the abstract communication channel, with true information as the input and prediction as the output (or vice versa, since the formula is symmetric).

This gets me back to the measurements of Forex Automaton prediction quality — the state of the art at the moment is the measurement of Pearson correlation coefficient between predicted and real value of logarithmic returns. I am eager to see what the plot will look like if Shannon’s mutual information is used instead of the correlation coefficient — this is particularly interesting, given Kelly’s interpretation of the quantity.

Where does this all leave the common theory with its Sharpe ratios and efficient portfolios? A recent (June 2009) article by Javier Estrada, Geometric Mean Maximization: an Overlooked Portfolio Approach? proved informative to me. Estrada contrasts Sharpe ratio maximization which leaves one with not one, but an entire efficient frontier of portfolios with geometric mean maximization (Kelly by other name) leading to the largest expected terminal wealth. Estrada notes that SRM (Sharpe ratio maximization) is a one-period framework, while GMM (geometric mean maximization) is a multi-period framework — I take this to mean that the biased nature of Sharpe when cumulative returns are concerned (as already noted, one of my early disappointments with this statistic) is thus acknowledged in the academic community.

Why do the practitioners overlook a useful criterion? — wonders Estrada. My guess is that the idea of Kelly’s “private wire” and the like is simply too alien to the version of the financial theory based on the postulates of market efficiency, thus making the corollary of Shannon’s information theory as applied to markets, the Kelly Criterion, genetically alien as well. One should never underestimate the role of the overall surrounding cultural and ideological context in the development of science. The symptomatic overstatement of symmetries by the theorists of “market efficiency” may be the case in point. Certainly, Kelly makes more sense to someone dealing with wealth generation on the basis of quantifiable and specific advantages, rather than just submitting oneself to the egalitarian random walk of the hypothetical “efficient market”. This is because Kelly addresses the problem of the value of information directly, it is at the heart of his approach.

Possible Figures Of Merit Related To Return On Investment: The Arithmetic And Geometric Mean

When evaluating the performance of a trading system, I calculate the first moment (an arithmetic mean of the series of returns) as well as the second one (a variance of the series). Originally my “Sharpe-like” ratio, used to adjust the return for the risk, was a ratio of the first moment to the square root of the second. The series of returns would be composed of annualized returns calculated every month.

However there is a subtlety. If {r1,r2,r3,… rn} is a series of monthly returns for some period (ratios of capital at the end of the period to that at the start of the period), then the total return for the same period will be the product r1r2r3… rn.

When each monthly return is already annualized (exponentiated to the 12th power), the opposite operation (applying power 1/12 to the product) is required to get the actual return for the year. This operation is known as taking the geometric mean.

Arithmetic mean is always greater than or equal to the geometric mean. This fact is well known from university math courses.

For example, if the series of returns is 1.0, 1.081, 0.9, 1.02, then the geometric average is 0.998084 and the arithmetic average is 1.00025. In this particular example, the arithmetic average of returns will tell you that you are making money whereas in fact you are losing.

Arithmetic mean equals the geometric mean in the particular case when all elements of the series are equal. The more volatility there is in the series, the more difference between the two means can be expected.

The bottom line is that the results of the research so far should be revisited using the simple return and the Sharpe ratio based on such return instead of the first moment of the series of annualized returns. The first moment gives a biased estimate of the actual return.

Nevertheless, since the main driving logic in the choice of parameters was minimization of the drawdown, the basic conclusions regarding the parameter choices are likely to stay the same.

<

Investment

“An investment operation is one which, upon thorough analysis promises safety of principal and an adequate return. Operations not meeting these requirements are speculative.” Apparently for Graham and Dodd, the term “speculation” has no positive connotations.

Pound Sterling (GBP) LIBOR Rates: Technical Predictability Overview

The original motivation for the technical, mostly correlation-based study of LIBORs was outlined in the USD LIBOR article. Like the USD, EUR and JPY LIBOR reports, this document begins with historical LIBOR charts for the Pound Sterling, continues with volatility analysis, and culminates with correlations of logarithmic returns in GBP LIBOR. You will see that predictable patterns in GBP LIBORs show great variation with loan term. Autocorrelations of short-term LIBORs look jittery on the next-day time scale. Autocorrelations of short term (s/n-o/n and 1-week) LIBOR exhibit the now familiar “bipolar disorder” pattern with the characteristic time period of no more than 2-3 days. The smooth wave-like patterns of intermediate term USD and EUR LIBORs, about 70 days in period, are also found in GBP. As the term duration increases, the main correlation pattern becomes that of predictive (non-zero time lag) positive correlation between different maturity terms as well as inside individual time series (autocorrelation).

LIBOR charts

History of s/n-o/n GBP LIBOR 2002-2008 History of 1 week GBP LIBOR 2002-2008 History of 1-month GBP LIBOR 2002-2008 History of 3-month GBP LIBOR 2002-2008 History of 6-month GBP LIBOR 2002-2008 History of 12-month GBP LIBOR 2002-2008

Fig.1: Historical GBP LIBOR rates charts, top to bottom: s/n-o/n, 1-week, 1-month, 3-month, 6-month and 12-month. Time axis is labeled in MM-YY format.

The most striking feature is the high volatility of s/n-o/n and 1-week LIBOR rates in 2002 which gradually goes down. For the short maturities, the markets jump the gun trying to anticipate the course of events almost regularly, to the extent this nervousness must represent a regular and significant speculative opportunity, if the market instruments tied to the LIBOR rates have the same features. Longer maturities develop patterns of their own while the shorter ones are dominated by the basic step-like pattern modulated by the short-range neurosis. This will be seen qunatitatively in the correlation plots.

 

LIBOR volatility

Table 1: Day-by-day volatilities (RMS) for the time series of logarithmic returns in GBP LIBOR in 2002-2008, various maturities

durationtime scalevolatility (RMS)
s/n-o/nday7.8×10-2
weekday2.6×10-2
monthday5.8×10-2
3 monthsday3.8×10-2
6 monthsday6.6×10-2
12 monthsday7.1×10-2

Volatility of GBP LIBOR seems to have no easy pattern in its dependence on duration term.

Distribution of logarithmic returns in s/n-o/n and 1-week GBP LIBOR rates Distribution of logarithmic returns in 1-month, 3-month and 12-month GBP LIBOR rates

Fig.2: Distributions of logarithmic returns in GBP LIBOR rates, top: s/n-o/n and 1-week, bottom: 1-month, 6-month and 12-month maturity. Volatility is a measure of the width of the return distribution.

The distribution of logarithmic returns on the day time scale looks rather complex, reflecting the evolution of the LIBOR pattern with time — the jittery picture of 2002 will certainly result in a different logarithmic return distribution than the 2006 and 2007. The core distributions may be power-law (remember that with returns already containing logarithm and with the vertical axis explicitly logarithmic, we are looking at what is effectively a log-log plot, where any power law dependence would have looked linear, with different power law exponents resulting in different slopes), but the long tails certainly do not belong to the same, if any, power law as the core distribution.

 

LIBOR autocorrelations

As with the Euro and the US Dollar LIBORs and with some of the most volatile forex exchange rates, the most prominent feature of the s/n-o/n and 1-week GBP LIBOR autocorrelations is the “bipolar disorder” pattern seen from the bins with large negative signal surrounding the zero-time lag bin. (The expression “bipolar disorder” in relation to the market is credited to Benjamin Graham). Continuing with the psychiatric analogy, these indicate rapid (next day or two, depending on LIBOR term) changes in the mood of the credit market, a price action followed by an immediate correction.

GBP s/n-o/n LIBOR autocorrelation, 1 day time scale GBP 1-week LIBOR autocorrelation, 1 day time scale GBP 1-month LIBOR autocorrelation, 1 day time scale GBP 3-month LIBOR autocorrelation, 1 day time scale GBP 6-month LIBOR autocorrelation, 1 day time scale GBP 12-month LIBOR autocorrelation, 1 day time scale

Fig.3: Autocorrelation of logarithmic returns in the historical GBP LIBOR is shown against the backdrop of statistical “noise”. The noise is obtained from martingale simulations based on the historical volatilities of LIBOR for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular time lag bin by analyzing 20 independent simulated pairs of uncorrelated time series. The RMS is a measure of accuracy in the determination of the correlation values, an irreducible uncertainty dependent on the amount of data and the time scale. Top to bottom: s/n-o/n, 1-week, 1-month, 3-month, 6-month, and 12-month data.

Fig.3 and subsequent figures ascertain the significance of the patterns by comparing with the statistical noise estimate, based on simulations devoid of correlations, but with volatility of the actual data. 1-week and longer term (but not 12-month) LIBOR autocorrelations are overall positive for the time lags of hunderds of days, with considerable evolution in shape. This is very different from forex exchange rates, and implies that in LIBOR, medium-range (several days) forecasting is straighforward for these maturities: betting on the continuation of a trend is the winning strategy. In other words, trend following is possible with LIBOR — forex exchange rates, on the contrary, generally justify no such strategy, and you will not find wide positive peaks in the forex return correlations.

The predictive positive zero-lag peak of 1-month and longer maturities has to be contrasted with the opposite feature seen in shorter maturities, namely the “bipolar disorder”, a tendency to form patterns where the strategy of betting on the trend reveral is more likely to succeed. This tendency shows up in the negative correlation magnitude at the lag that corresponds to the time it takes for the trend reversal. In GBP LIBOR, s/n-o/n and 1-week data, the time is no more than 2-3 days. Trend following is not a viable strategy with s/n-o/n and 1-week LIBOR: here, betting on the next-day trend reversal or using longer range correlations, some of which just as sharp, seems to be the surest strategy. 1-month, 3-month and 6-month figures show oscillations with what looks like 70 to 75 day period (counting business days only). Not sure what this has to do with periodicity of BOE meetings — but 70 days is twice the FOMC’s regular period. In fact, similar periodicity has been seen in the USD and EUR LIBOR autocorrelations for comparable maturities.

 

Cross-correlations of LIBOR terms

Next, I am going to look at correlation between LIBOR rates of different maturities for various time lags. These help answer the question to what extent one LIBOR term can be predicted on the basis of any others. Here is the summary, followed by the data.

The covariance of different maturity terms (amplitude of the zero time-lag peak) is seen to go down as the difference in maturities grows; similar maturities are correlated tighter. The correlations become overall more positive between longer-term maturities.

Correlations between s/n-o/n and longer term LIBOR rates

Correlation between logarithmic returns in s/n-o/n and 1-week GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 1-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 3-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 6-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 12-month GBP LIBOR rates as a function of time lag, days

Fig.4: Correlation between logarithmic returns in s/n-o/n and, top to bottom: 1-week, 1-month, 3-month, 6-month and 12-month GBP LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

 

Correlations between 1-week and longer term LIBOR rates

Correlation between logarithmic returns in 1-week and 1-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-week and 3-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-week and 6-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-week and 12-month GBP LIBOR rates as a function of time lag, days

Fig.5: Correlation between logarithmic returns in 1-week and, top to bottom: 1-month, 3-month, 6-month and 12-month GBP LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

 

Correlations between 1-month and longer term LIBOR rates

Correlation between logarithmic returns in 1-month and 3-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-month and 6-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-month and 12-month GBP LIBOR rates as a function of time lag, days

Fig.6: Correlation between logarithmic returns in 1-month and, top to bottom: 3-month, 6-month and 12-month GBP LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

 

Correlations between 3-month and longer term LIBOR rates

Correlation between logarithmic returns in 3-month and 6-month GBP LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 3-month and 12-month GBP LIBOR rates as a function of time lag, days

Fig.7: Correlation between logarithmic returns in 3-month and, top to bottom: 6-month and 12-month GBP LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

What Do You Mean By “Predictable” Or “Predictability” When You Talk About Forex?

Imagine tossing a coin which is slightly bent in a way which is known to you. The bend is almost unnoticeable but it does exist and this fact will become obvious after a long enough series of coin tosses, if you do the book-keeping accurately. Trading the markets with a good trading system is similar. The amount of predictability is small, making it justifiable for people who do not have access to an analysis technique of sufficient sensitivity to talk about “efficient markets” — the markets indeed look “efficient” to their methods of observation. The fact of predictability only becomes undeniable after hundreds or thousands of “coin tosses” (trades).

The world economy is never in equilibrium. “I assume that markets are always wrong” said George Soros. A market actor (investor, trader, central banker) requires finite time to analyze the changing environment and to make a decision. Yet more time and resources are needed to turn it into action visible by the market. No analysis is perfect, and some are less perfect than others. In addition, people who share common educational and cultural background tend to make similar, rather than the most efficient, decisions. Groupthink exists inside communities and large organizations. For these and other reasons, the so-called market inefficiency exists. Once the market inefficiency is quantified, predictive power (and a winning strategy) is only a step away, since such inefficiency — always specific and quantifiable — is in some sense but a deviation from the perfect symmetry which random noise (aka “efficient market”) would possess.

US Dollar (USD) LIBOR Rates: Technical Predictability Overview

This article begins a series of analysis reports investigating a degree of predictability in the LIBOR rates, a popular capital cost indicator. The analysis is based on historical LIBOR interest rate data released by the British Bankers Association. I continue with the same technique proved useful in the predictability analysis of forex exchange rates, as our interest in the interest rates in general is in part provoked by the results of the latter analysis, namely:

  • sometimes, one forex exchage rate can “show the way” to a number of others, or in other words, foretell (in a probabilistic or statistical sense) their movement.
  • when that happens, it is usually the exchange rate with a large interest rate differential showing the way to the ones with lower interest rate differentials.

Autocorrelations in s/n-o/n USD LIBOR 2002-2008

Fig.1: LIBOR heartbeat: autocorrelation in logarithmic returns of historical USD LIBOR rates, s/n-o/n duration, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the recorded LIBOR volatility for the period under study (2002-2008). The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series. The LIBOR shows strong regular structure with a period of 10 business days (two weeks). Time lag is measured in days. The familiar jump-the-gun pattern (strong negative signal around zero time lag) seen sometimes in forex, is also visible here. This is the level of predictability one can only dream of in forex exchange rates, yet it is the interest rates that drive forex. Is LIBOR always that predictable?

Obviously, when exploring these “loopholes” or market inefficiencies for wealth generation, an algorithmic trader or a forex trading system (an automated decision making algorithm such as the one being built here on Forex Automaton™ site) must be mindful of the picture of LIBOR rates and its evolution, albeit in a somewhat different context than a long-term money manager. Being able to predict events, even in a weak statistical sense, is even better than merely following. Besides being useful via their implications for forex forecasting, LIBORs form an underlying indicator for derivatives of their own. LIBOR futures contracts and options on such contracts are traded on the CME. How does the predictability of LIBORs compare with that of currencies? Which one, LIBOR or forex, is more attractive to trade? Answering these questions, or providing a technical analysis framework to approach the answers, while leaving the fundamentals and event-driven trends aside, this series of articles about correlation features in LIBORs will serve as a useful compliment to our set of forex correlation analysis notes. I start this new series of articles with the all-important US Dollar LIBOR.

Executive Summary

The time series of US Dollar LIBORs is highly volatile; on the day scale, the distribution of “returns” (daily increments) is not only not lognormal but for some maturities not even power law. The main correlation pattern is positive correlation, between different terms as well as inside individual time series (autocorrelation). The heart-beat pattern of short-range LIBOR is replaced by longer-range waves and finally disappears in 12-month term data.

LIBOR charts

BBA tracks LIBORs for various loan durations (maturities). I focus on spot-next/overnight (s/n-o/n), 1 week, 1 month, 3 month, and 1 year maturities.

History of s/n-o/n USD LIBOR 2002-2008 History of 1 week USD LIBOR 2002-2008 History of 3-month USD LIBOR 2002-2008 History of 12-month USD LIBOR 2002-2008

Fig.2: Historical USD LIBOR rates charts, top to bottom: s/n-o/n, 1-week, 1-month, 3-month and 12-month. Time axis is labeled in MM-YY format.

Comparing the histories of different LIBOR maturities in different panels of Fig.2, you see their regular and sharp pattern becoming progressively less sharp as the maturity increases, with the shortest, s/n-o/n LIBOR, having the sharpest structure. The origin of the negative spikes surrounding the zero time-lag peak is quite clearly visible: the history of LIBOR is full of false alarm events when the rate suddenly changes its course (typically, jumps up) only to correct itself completely the next day. The money markets jump the gun trying to anticipate the course of events almost regularly, to the extent this nervousness must represent a regular and significant arbitrage opportunity, if the market instruments tied to the LIBOR rates have the same features.

LIBOR volatilities

Table 1: Day-by-day volatilities (RMS) for the time series of logarithmic returns in USD LIBOR in 2002-2008.

durationtime scalevolatility (RMS)
s/n-o/nday3.1×10-2
weekday1.4×10-2
monthday8.5×10-3
3 monthsday8.4×10-3
12 monthsday1.9×10-2

Distribution of logarithmic returns in s/n-o/n and 1-week USD LIBOR rates Distribution of logarithmic returns in 1-month, 3-month and 12-month USD LIBOR rates

Fig.3: Distributions of logarithmic returns in USD LIBOR rates, top: s/n-o/n and 1-week, bottom: 1-month, 3-month and 12-month maturity. Volatility is a measure of the width of the return distribution.

As you could have guessed from inspecting Fig.2, s/n-o/n LIBOR is more volatile than 1-week, whereas 12-month LIBOR is more volatile than 1-month and 3-month. Forex distributions on a shorter scale (one hour) look much less volatile than LIBOR, the “fat tails” in LIBOR distributions are truly amazing.


LIBOR autocorrelations

USD 1-week LIBOR autocorrelation, 1 day time scale USD 1-month LIBOR autocorrelation, 1 day time scale USD 3-month LIBOR autocorrelation, 1 day time scale USD 12-month LIBOR autocorrelation, 1 day time scale

Fig.4: Autocorrelation of logarithmic returns in the historical USD LIBOR is shown against the backdrop of statistical “noise”. The noise is obtained from martingale simulations based on the historical volatilities of LIBOR for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular time lag bin by analyzing 20 independent simulated pairs of uncorrelated time series. The RMS is a measure of accuracy in the determination of the correlation values, an irreducible uncertainty dependent on the amount of data and the time scale. Top to bottom: 1-week, 1-month, 3-month, and 12-month data.

As was seen in Fig.1, the pediodic “heartbeat” pattern in the overnight LIBOR autocorrelation shows no signs of abaiting for the range of lags as long as 200 days. In Fig.4, you see that this “heartbeat” is unique to the overnight LIBOR, while some “gun-jumping” (short range negativeness) is also seen in 1 week LIBOR loan term data, but is gone in longer terms. Fig.1 and Fig.4 ascertain the significance of the patterns by comparing with the statistical noise estimate, based on simulations devoid of correlations, but with volatility of the actual data. 1-week, 1-month and 3-month LIBOR autocorrelations are overall positive, with a strong short-range peak. This is very different from forex exchange rates, and implies that medium-range LIBOR forecasting is straighforward for these duration terms: betting on the continuation of a trend, no matter whether the trend is up or down, is the winning strategy. In other words, trend following is possible with LIBOR — forex exchange rates, on the contrary, usually justify no such strategy, and you will not find wide positive correlation peaks in forex. Neither does the 12-month LIBOR where the wide peak around zero time lag is gone.

Federal Open Markets Committee (FOMC) holds its regular meetings every 7 weeks, that is 35 business days. One might expect that regularity to result in a feature of some kind corresponding to this time scale. The wavelength of the 1-month and 3-month LIBOR oscillations, see Fig.4, which is about 70 days, may be linked to that, being a multiple of 35.

Cross-correlations of LIBOR terms

Next, I am going to look at correlation between LIBOR rates of different maturities for various time lags.

Correlation between logarithmic returns in s/n-o/n and 1-week USD LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 1-month USD LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 3-month USD LIBOR rates as a function of time lag, days Correlation between logarithmic returns in s/n-o/n and 12-month USD LIBOR rates as a function of time lag, days

Fig.5: Correlation between logarithmic returns in s/n-o/n and, top to bottom: 1-week, 1-month, 3-month and 12-month USD LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

Correlation between logarithmic returns in 1-week and 1-month USD LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-week and 3-month USD LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-week and 12-month USD LIBOR rates as a function of time lag, days

Fig.6: Correlation between logarithmic returns in 1-week and, top to bottom: 1-month, 3-month and 12-month USD LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

Correlation between logarithmic returns in 1-month and 3-month USD LIBOR rates as a function of time lag, days Correlation between logarithmic returns in 1-month and 12-month USD LIBOR rates as a function of time lag, days

Fig.7: Correlation between logarithmic returns in 1-month and, top to bottom: 3-month and 12-month USD LIBOR rates as a function of time lag, days, shown against the backdrop of statistical noise (red). The noise is obtained from martingale simulations based on the historical LIBOR volatilities for the period under study. The noise is presented as mean plus-minus 1 RMS, where RMS characterizes the distribution of the correlation value obtained for each particular bin by analyzing 20 independent simulated pairs of uncorrelated time series.

Naturally, individual LIBOR maturities are generally positively correlated with one another. Speaking of forecasting, more often than not we see more positive correlation in the positive range of time lag. All correlated pairs are ordered so that shorter term is first, longer term second. The time lag being

t1-t2,

the conclusion is that often, shorter term follows and longer term leads. 12-month term is seen to lead 1-week and 1-month; 3-month term is seen to lead s/n-o/n; so is 1-month.