4 views

Uploaded by sumitbeniwal

Research paper

save

- 79315_Lecture 6 - Qualitative and Quantitative Sampling
- Impact of Random Failures and Attacks on Poisson and Power-Law Random Networks
- BESO2D Command Line Usage
- PARAMETRICS AND OPTIMIZATION USING ANSOFT HFSS_10.1.1.169.1598[1]
- BI Integrated Planning - Scenario
- RF Power Control and Handover Algorithm_ Variable Averaging Window Size
- WCDMA Prfile 16
- Cap05-02
- Fianncial Markets
- Basic Creep
- 108.pdf
- Pareto Law in a Kinetic Model of Market With Random Saving Propensity
- 40 Multiple Choice Questions in Basic Statistics
- Bridge Class
- ST06
- Exam 050272RR Probability a Breeder Records Probabilities for Two Variables in a Population Answer
- BA201 Engineering Mathematic UNIT2 - Measures of Central Tendency
- A New Curve for Fire Temperature in Compartiment Fire
- Add maths questions
- Self Organized Criticality and Phase Locking in Models of Production and Inventory Dynamics
- Chap006 New
- Acoustical Monitoring of Fish Density, Behavior, And Growth Rate in a Tank(1)
- Tutorial 04
- 100 Prisoners Lightbulb
- Test for Normality
- D800010X062
- Axiomatic o Ejemplo
- Bayesian Analysis Sample
- BMS MAROS
- Creating Product and Process Control Systems
- Sapiens: A Brief History of Humankind
- The Unwinding: An Inner History of the New America
- Yes Please
- Dispatches from Pluto: Lost and Found in the Mississippi Delta
- Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future
- The Innovators: How a Group of Hackers, Geniuses, and Geeks Created the Digital Revolution
- John Adams
- Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America
- A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story
- Grand Pursuit: The Story of Economic Genius
- This Changes Everything: Capitalism vs. The Climate
- The Emperor of All Maladies: A Biography of Cancer
- The Prize: The Epic Quest for Oil, Money & Power
- Team of Rivals: The Political Genius of Abraham Lincoln
- The New Confessions of an Economic Hit Man
- The World Is Flat 3.0: A Brief History of the Twenty-first Century
- Rise of ISIS: A Threat We Can't Ignore
- Smart People Should Build Things: How to Restore Our Culture of Achievement, Build a Path for Entrepreneurs, and Create New Jobs in America
- The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers
- How To Win Friends and Influence People
- Angela's Ashes: A Memoir
- Steve Jobs
- Bad Feminist: Essays
- The Light Between Oceans: A Novel
- Leaving Berlin: A Novel
- The Silver Linings Playbook: A Novel
- The Sympathizer: A Novel (Pulitzer Prize for Fiction)
- Extremely Loud and Incredibly Close: A Novel
- You Too Can Have a Body Like Mine: A Novel
- The Incarnations: A Novel
- Life of Pi
- The Love Affairs of Nathaniel P.: A Novel
- A Man Called Ove: A Novel
- Bel Canto
- The Master
- We Are Not Ourselves: A Novel
- The First Bad Man: A Novel
- The Rosie Project: A Novel
- The Blazing World: A Novel
- Brooklyn: A Novel
- The Flamethrowers: A Novel
- The Art of Racing in the Rain: A Novel
- Wolf Hall: A Novel
- The Wallcreeper
- Interpreter of Maladies
- Beautiful Ruins: A Novel
- The Kitchen House: A Novel
- The Perks of Being a Wallflower
- The Bonfire of the Vanities: A Novel
- Lovers at the Chameleon Club, Paris 1932: A Novel
- A Prayer for Owen Meany: A Novel
- The Cider House Rules
- Mislaid: A Novel

net/publication/46462361

**Patterns in high-frequency FX data: Discovery of
**

12 empirical scaling laws

Article in Quantitative Finance · January 2008

DOI: 10.1080/14697688.2010.481632 · Source: RePEc

CITATIONS

READS

23

63

3 authors, including:

Alexandre Dupuis

Richard Olsen

University of Geneva

Olsen Ltd

48 PUBLICATIONS 1,082 CITATIONS

36 PUBLICATIONS 2,580 CITATIONS

SEE PROFILE

**All in-text references underlined in blue are linked to publications on ResearchGate,
**

letting you access and read them immediately.

SEE PROFILE

**Available from: Richard Olsen
**

Retrieved on: 20 September 2016

arXiv:0809.1040v2 [q-fin.ST] 22 Jun 2010

**Patterns in high-frequency FX data:
**

Discovery of 12 empirical scaling laws

J.B. Glattfelder∗†, A. Dupuis†‡ and R.B. Olsen†‡

† Olsen Ltd., Seefeldstrasse 233, 8008 Zurich, Switzerland

‡ Centre for Computational Finance and Economic Agents (CCFEA), University of Essex, UK

Abstract

We have discovered 12 independent new empirical scaling laws in foreign exchange

data-series that hold for close to three orders of magnitude and across 13 currency exchange rates. Our statistical analysis crucially depends on an event-based approach

that measures the relationship between different types of events. The scaling laws

give an accurate estimation of the length of the price-curve coastline, which turns out

to be surprisingly long. The new laws substantially extend the catalogue of stylised

facts and sharply constrain the space of possible theoretical explanations of the market mechanisms.

Keywords: Scaling Laws; High-Frequency Finance; Foreign Exchange; Time-Series

Analysis; Gaussian Random Walk Models

1

Introduction

The global financial system has recently been rocked by losses that could total four trillion USD (International Monetary Fund [2009]). The crisis is seriously undermining the

functioning of the financial system, the backbone of the global economy. This suggests an

acute deficiency in our understanding of how markets work. Are there “laws of nature”

to be discovered in financial systems, giving us new insights? We approach this question

by identifying key empirical patterns, namely scaling-law relations. We believe that these

universal laws have the potential to significantly enhance our understanding of the markets.

Scaling laws establish invariance of scale and play an important role in describing

complex systems (e.g. West et al. [1997], Barab´asi and Albert [1999], Newman [2005]).

In finance, there is one scaling law that has been widely reported (M¨

uller et al. [1990],

Mantegna and Stanley [1995], Galluccio et al. [1997], Guillaume et al. [1997], Ballocchi et al.

[1999], Dacorogna et al. [2001], Corsi et al. [2001], Di Matteo et al. [2005]): the size of the

∗

Corresponding author. Email: jbg@olsen.ch.

1

roughly being the sum of all price moves of a given threshold. we analyse the price data of the foreign exchange (FX) market. [1996]. Joulin et al. But how many times longer? The scaling laws described in this paper provide a surprisingly accurate estimate and highlight the importance of not only considering tail events (Sornette [2002]). Cχ (p) are the scaling-law parameters. Sornette [2000]. The 2 . a complex network made of interacting agents: corporations. Gabaix et al. In section 3 the methods and the data are described and we conclude with some final remarks in section 4. The coastline. but set these in perspective with the remarkably long coastline of price changes preceding them. over-the-counter market that is not constrained by specific exchange-based rules. Barndorff-Nielsen and Prause [2001]. With a daily turnover of more than 3 trillion USD (Bank for International Settlements [2007]) and with price changes nearly every second. Our main results are presented in section 2. Farmer and Lillo [2004]. In this study we consider five years of tick-by-tick data for 13 exchange rates through November 2007 (see section 3.1 The laws and beyond The new scaling laws Interest in scaling relations in FX data was sparked in 1990 by a seminal paper relating the mean absolute change of the logarithmic mid-prices. Finally. 2 2. then cross-check our results by establishing quantitative relations amongst them and discuss the coastline. In the challenge of identifying new scaling laws. The remainder of the paper is organised as follows. since the seminal work of Mandelbrot [1963] we know about the fractal nature of price curves. institutional and retail traders.. who themselves form an intricate web of interdependence.1 for a description of the data set). We start by enumerating the empirical scaling laws. at fine levels of resolution may be far longer than one might intuitively think. and Eχ (p). It is a common occurrence for an exchange rate to move by 10 to 20% within a year. (0a) where ∆χi = χi − χi−1 and χi = χ(ti ) = (ln bidi + ln aski )/2 is the logarithmic midprice of a currency pair at time ti . appendix A contains tables with all the estimated scaling-law parameters. the FX market offers a unique opportunity to analyse the functioning of a highly liquid. Lux [2006].g. Di Matteo [2007]) even though there has been no consensus amongst researchers for why the scaling law exists (e.average absolute price change (return) is scale-invariant to the time interval of its occurrence. to the size of the time interval (M¨ uller et al. This scaling law has been applied to risk management and volatility modelling (see Ghashghaie et al. However. and brokers trading through market makers. [1990]) h|∆χ|ip = ∆t Cχ (p) Eχ (p) . [2003]. It should be noted that our study is not related to the analysis of lead-lag relationships. [2008]). Bouchaud [2001]. sampled at time intervals ∆t over a sample of size n∆t.

relating the number N(∆χdc ) of so-called directional changes to the the directional-change sizes ∆χdc E ∆χdc N. Here we confirm laws (0a) and (0b) considering xi (see figures 3a – c). This requires a time interpolation scheme (described in section 3. The dissection algorithm measures occurrences of a price change ∆xdc from the last high or low (i. Figure 1 depicts how the price curve is dissected into so-called directionalchange and overshoot sections. i. Appendix A provides tables of the estimated parameter values for all the laws and for the 13 exchange rates as well as for a Gaussian random walk (GRW) model. In detail. χEUR-USD = −χUSD-EUR ) it is i i more natural as. the fixed event thresholds of different sizes define focal points.1.dc N(∆χdc ) = . the overshoot associated with the previous directional change is determined as the difference between the price level at which the last directional change occurred and the extrema. In section 3. 2}. stable patterns of scaling and reduces the level of complexity of real-world time series. described in section 3.e. [1997]. at news announcements there are spurts of market activity. extrema). every table lists the average parameter values over all 13 currency pairs and their sample standard deviations. Later. Throughout the paper we consider a simpler definition of the price given by xi = (bidi + aski )/2 where price moves are defined as ∆xi = (xi − xi−1 )/xi−1 . a second scaling law was reported by Guillaume et al. At each occurrence of a directional change. the flow of time is discontinuous: over weekends trading comes to a standstill or. where patterns emerge for successions of events at different magnitudes.1) which we will also employ when necessary in the following. respectively. The high and low price levels are then reset to the current price and the mode alternates.2 the pseudocode for the directional-change count is provided in algorithm 2. and report on 12 new independent scaling laws holding across 13 exchange rates and for close to three orders of magnitude. if it is in an up or down mode. considering either χi or xi leads to very similar results even for large spread values. whereas xi is the arithmetic mean. blurring out irrelevant details of the price evolution. Extending this event-driven paradigm further enables us to observe new. Note that for law (0a) the data is sampled at fixed time intervals ti = i∆t. percentages are more intuitive to manipulate than differences between logarithmic values. in 1997.. constituting an event-driven approach. Although the definition of xi looses the mathematical feature of χi of behaving anti-symmetrically under price inversions (e.g.e.dc In financial markets. the high when in up mode or low when in down mode. the confinement of analysing returns as observed in physical time is overly restrictive. However. and p is omitted if equal to one. inversely. Law (0b) is a first attempt at establishing a new paradigm by looking beyond such constraints within financial data. Table 1 shows the 3 . usually with p ∈ {1.. practically. (0b) CN. 1/p P averaging operator is hxip = 1/n nj=1 xpj . Note that the logarithmic mid-price √ is equivalent to the logarithm of the geometric mean of the bid and ask prices χi = ln bidi aski . In law (0a).. This alternative approach defines an activity-based time-scale called intrinsic time. In addition.

7%. We start the enumeration of the laws by a generalisation of equation (0b) that relates the average number of ticks observed during a price move of ∆x to the size of this threshold EN.2.02%. (1) CN. The second law counts the average yearly number N(∆x) of price moves of size ∆x EN.tck where a tick is defined as a price move larger than (in absolute value) ∆xtck = 0.23%. (b) zoomed-in 36 hour price sample onto a reduced set of so-called directional-change events defined by a threshold (a) ∆xdc = 1.x The computation of this law is provided in algorithm 1.x ∆x . We annualise the number of observations of laws (0b) and (2) by dividing them by 5. (2) N(∆x) = CN. Note the increase of spread size during the two weekends with no price activity. These directional-change events (diamonds) act as natural dissection points. In essence. Time scales depict physical time ticking evenly across different price-curve activity regimes. The directional-change computation is detailed in algorithm 2 of section 3. (b) ∆xdc = 0. however. be altered without destroying the scaling-law relation. decomposing a total-price move between two extremal price levels (bullets) into so-called directional-change (solid lines) and overshoot (dashed lines) sections. whereas intrinsic time triggers only at directional-change events.tck ∆x hN(∆xtck )i = . Law (1) is plotted in figure 2. The definition of a tick can. independent of the notion of physical time. estimated scaling-law parameters for EUR-USD. the number of years in our data 4 . this law counts the average number of ticks observed during every price move ∆x.Figure 1: Projection of a (a) two-week.

1% and 3. of ticks) AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2e-5 4e-5 101 6e-5 102 0 10−1 0 100 Average number of ticks 105 Tick-count scaling law 0 10-2 20000 40000 100 10−1 60000 80000 101 102 ∆x (%) Figure 2: Scaling law (1) is plotted where the x-axis shows the price move thresholds of the observations and the y-axis the average tick numbers. Insets show the distribution of the EUR-USD observations (drawn above their x-axis) for selected threshold values of 0.0%. See appendix A for the values of the estimated scaling-law parameters. defined as the difference between the high and low price levels. We have also discovered laws relating the time during which events happen to the magnitude of these events. (3) Cmax (p) where ∆xmax = max{x(τ ). sample. The solid line shows the raw data for EUR-USD. during a time interval ∆t. For the remaining 12 currency pairs and the Gaussian random walk benchmark model the raw data is displayed with dots.x h∆tx i = . number of ticks) 200 400 600 800 Kernel density estimation ∆x = 3. Law (4) relates the average time interval h∆tx i for a price change of size ∆x to occur to the size of the threshold E ∆x t.1% (Density vs.0% (Density vs. to the size of that time interval Emax (p) ∆t h∆xmax ip = .104 2e-2 103 0 4e-2 Kernel density estimation ∆x = 0. no. A tick is defined as a price move of 0. t]} − min{x(τ ). τ ∈ [t − ∆t. The next scaling law relates the average maximal price range ∆xmax . τ ∈ [t − ∆t. t]} and (3) holds for p = 1. (4) Ct.02%.x 5 . Law (2) and all the following scaling laws are given in figure 3. 2.

The raw data is plotted for the 13 currency pairs with dots and for the Gaussian random walk model with dashes. ∗ = os t 104 hN(∆xdc tck )i 100 10−1 ∆xdc (%) 100 101 102 103 104 105 106 N(∆x) h∆tdc i (s) Total-move tick count Law (11). ∗ = to 100 10−1 ∆xdc (%) 10−2 Directional-change tick count p 104 Cumulative directional change Law (12). ∗ = os 103 101 102 103 104 105 106 107 100 r 104 105 ∆t (s) 105 Overshoot move Law (9). ∗ = dc 101 ∆xdc coast (%) 100 10−1 ∆xdc (%) 100 100 10−1 ∆xdc (%) 10−2 Overshoot tick count Law (11). See appendix A for the values of the estimated scaling-law parameters. p = 1 m 106 ∆xos coast (%) f 104 105 ∆t (s) 100 Maximum price move e 103 102 102 102 N(∆xdc ) 106 Price move count Law (2) d Law (0b) 102 104 105 ∆t (s) 10−1 100 101 103 Directional change count 101 101 100 h∆xi2 (%) 10−2 102 10−2 c Law (0a). p = 2 10−1 h∆xto i (%) 102 10−2 10−2 Quadratic mean max price move Law (3). Symbols are as in figure 2. p = 1 100 101 102 103 104 105 106 Mean price move a 100 10−1 ∆xdc (%) 10−2 100 10−1 ∆xdc (%) Figure 3: Plots of all scaling laws described in the text. ∗ = to Law (11). ∗ = os s 100 10−1 ∆xdc (%) 10−2 100 10−1 ∆xdc (%) Coastline (cum. ∗ = to 100 Total price move 102 106 100 10−1 ∆xdc (%) Law (4) 101 101 100 h∆xmax i (%) 10−1 10−2 104 105 ∆t (s) 101 103 i q g Law (3). ∗ = dc 10−2 10−1 10−2 h∆tto i (s) j 101 102 103 104 105 106 107 Law (9). 6 . total move) 102 103 104 Law (12).b h∆tos i (s) 10−1 100 10−1 ∆xdc (%) 10−2 100 10−1 ∆xdc (%) 10−2 Time of directional change o l 10−2 Time of overshoot Law (10). ∗ = dc 10−2 10−1 100 101 102 103 104 Law (10). ∗ = os h 104 10−2 10−2 Mean time of price move 103 h∆tdc i (s) 100 h∆xdc i (%) 10−1 10−2 100 10−1 ∆xdc (%) 100 101 102 103 104 105 106 107 n 103 Number of ticks overshoot 10−2 Directional-change move 10−2 h∆xos i (%) 100 h∆xi2 (%) 100 10−1 ∆xdc (%) Law (9). p = 2 10−1 101 100 10−1 10−2 h|∆x|i1 (%) Quadratic mean price move b Law (0a). ∗ = to 10−1 ∆x (%) Law (5) 103 10−2 100 Time during directional changes 101 102 103 104 105 106 107 107 105 h∆tx i1 (s) 103 101 106 10−1 ∆x (%) 100 10−1 ∆xdc (%) Cumulative overshoot Law (12). ∗ = dc 10−2 ∆xcoast (%) 102 103 104 105 10−1 100 101 Number of ticks total k 100 101 Time of total move Law (10).

∗ E ∆xdc t. Considering cumulative price moves instead of the averages in laws (9) leads to another triplet of laws Ecum.dc h∆tdc i = .∗ E ∆xdc N. which themselves decompose into directional-change (DC) and overshoot (OS) parts.99 and Cx. The average parameter values (across the 13 currency pairs given in appendix A) of law (9) display a peculiar feature: on average.∗ ∗ .5 2∆x + 0.∗ ∗ h|∆x |i = . (6) (7) (8) This decomposition leads to nine additional scaling laws.to ≈ 0. as seen in table A11. This result is also found by computing the probable of the price within a binomial tree as Pn path 2 3 i n→∞ 0.04 and Cx. as the impact of the effect of a tick exceeding the exact threshold systematically overestimates h|∆xdc |i (see figure 3m).∗ where ∗ stands for {tm. The actual deviation to Ex. = ∆x i i 0.5 3∆x + . os}. and number of ticks can then be written as h|∆xtm |i = h|∆xdc |i + h|∆xos |i. dc os hN(∆xtm tck )i = hN(∆xtck )i + hN(∆xtck )i. dc. Note that h|∆xdc |i = ∆xdc holds by construction.dc (5) Thus laws (4) and (5) relate the average numbers of seconds that elapse between consecutive price moves or directional changes. making the total move double the size of av av the directional-change threshold h|∆xtm |i ≈ 2∆xdc (Ex. h∆ttm i = h∆tdc i + h∆tos i. . .51). it contains roughly twice as many ticks and takes twice as long to unfold. considering directional changes of threshold ∆xdc E ∆xdc t. So although in terms of size the overshoot price move is approximately as big as the direction-change threshold.dc = 1 and Cx. respectively.os ≈ 1. (12) C cum.∗ ∗ hN(∆xtck )i = .5 ∆x + 0.∗ n X ∆xdc ∗ ∗ ∆xcum = |∆xi | = . (11) CN. is given by the increasing noise for small thresholds.to ≈ 0.os ≈ 1.06).dc = 1.∗ i=1 7 . waiting time. The total price move. Next we unveil a set of scaling laws emerging from the identification of directionalchange events (see figure 1 and algorithm 2) that make up the so-called total-move (TM) segments. Ct. (9) Cx. a directional change ∆xdc is followed by an overshoot of the same magnitude av av h|∆xos |i ≈ ∆xdc (Ex.and similarly.5 −−−→ 2∆x. A similar feature holds for dc the waiting times and number of ticks: h|∆tos |i ≈ 2h|∆tdc |i and hN(∆xos tck )i ≈ 2hN(∆xtck )i. (10) h∆t i = Ct. where the average values are functions of the directional-change thresholds ∆xdc E ∆xdc x.

It should be noted however. that this is only the case for law (0a) with p = 2. 2. To cross-check the laws we first compare laws (2) and (4). j. We show that not only consistency requirements link the various laws. In the next section. However. (13) where Y is the number of seconds in a year. and omit the trivial scaling law h|∆xdc |i ∝ ∆xdc . where Y = 31′ 553′ 280 seconds. Indeed. it is striking to observe how close this simple model can be to the average properties of the market data.93 = −EN. Our results show that most of the currency pairs exhibit similar average behaviour. We employ unitary intervals ∆t of one second (see 3.x and Ct.500 (see table A5). This implies that Et.x .This concludes the presentation of 17 new scaling laws: we count equation (3) twice for p = 1. where the exponent of 1/2 can be analytically derived for a GRW. σ 2). Mandelbrot and Hudson [2004]. and use the fact that the average price-move time equals the sample length divided by the number of observations h∆tx i = Y /N(∆x). 2.x . b. as they are related to laws sensitive to the time-spacing of ticks.x ↔ −EN. It is well-known that the statistical properties of a GRW are different from the ones observed in empirical data.x CN. that the additional deviations between the empirical data and the GRW (seen in figures 3a.x = 1. 8 . it is incorrect to assume an exponent of 1/2 (or 2) for the other laws. h. It is often claimed that scaling-laws with exponent 1/2 are expected for a GRW (Guillaume et al. (14) The estimated EUR-USD parameters from table 1 allow us to verify equation (14) as Et. It should also be noted that any realisation of a GRW has an additional degree of freedom next to the specification of the price-move distribution ∆xi ∼ N (µ. Notable differences are seen in law (3) (see figures 3e and f) which reveals an unintuitive result: the bell-curve distribution of price moves leads to an average maximal price move that is roughly eight times larger than observed for the empirical data. we will address the question of how many of these laws are independent and hence can be understood as primary laws.x = 1. but that they can be combined to yield new scaling-law relations. and r) are an artifact of this simplistic choice.x ↔ Y 1/EN. [1997]) and hence observing this in empirical data would comply with the efficient market hypothesis (Fama [1970]). Furthermore.1).457. g. Yet for this law the real market data show an exponent averaging around 0. This however does not appear to be true in most of the laws for EUR-CHF as seen in figure 3. n.2 Exploring the space of scaling laws The scaling laws do not represent isolated patterns but are in fact related to each other.x and Ct. The value we actually measure is indeed 0. It could therefore be argued. namely the distribution of the time intervals ∆t. our realisations of a GRW show exponents for all the laws but (0a) for p = 2 to actually differ from 1/2 (and 2).23 · 10−3 = Y 1/EN.x CN.

The discrepancies seen in the consistency checks are in line with the fitting errors (see appendix A) and give us confidence to further proceed in exploring the space of scaling laws.23 · 10−3 ≈ 1.2129 = 1.94 −0. equivalent relations hold between laws (0b) and (5): Et.91 1.dc = 1.2 · 10−2 2.x = 1.49 1.93 −1.0 · 102 1.8 · 101 1.98 1.88 0.5 · 102 8.28 · 10−3 = Cx (1)−Ex (1) .98 −0. (13) that must hold for the cumulative and dissected average moves h|∆x∗ |i = ∆x∗cum /N(∆xdc ).1 · 102 Similarly.4 = ∆xtm cum /N(∆xdc ). overshoot Equation 1 2 3 (p = 1) 3 (p = 2) 4 5 9 9 10 10 10 11 11 11 12 12 12 12 Table A2 A3 A6 A7 A8 A9 A10 A12 A13 A14 A15 A16 A17 A18 A19 A20 A21 A22 E 1.4%.2132 ≈ 0. ∆xtm cum = ∆xcum + ∆xcum agrees to within an error of 0. We find for the total move.87 −0.05 · 10−3 ≈ 1.89 2.9 · 10−1 9.9 · 10−1 1.93 ≈ 2.x ↔ Cx (1)−Ex (1) . os}.6 · 10−3 1.9 · 10−2 4.x = 1. The scaling laws can be assembled to produce additional laws.5 · 100 1.93 1. directional change Cumul.5%.02 1.4 · 10−3 1. considering the directional-change and the overshoot part redc os sults in errors smaller than 0.1 · 10−2 9.9 · 105 1. Furthermore.01 = 1/Ex (1) and Ct. total move Cumul. laws 9 .dc = 1.92 C 2.dc .total move w. As an example. (15) We indeed have Et. 244.2/5.52 0. dc.88 ≈ 1. In addition.2 · 10−3 1.3 · 105 1.95 −0. implying Et.1%: h|∆xtm |i = 0.x ↔ −Ex−1 (1) and Ct.Table 1: Estimated scaling law parameter values considering EUR-USD.0 1. Name Tick count Price move count Maximum price move Maximum price move Time of price move Time of directional change Total price move Overshoot move Time of total move Time of directional change Time of overshoot Total-move tick count Directional-change tick count Overshoot tick count Cumul. considering EURUSD.89 1.3 · 10−2 2. 843. with a threshold of ∆xdc = 0.1 · 10−3 4. costs Cumul.93 0. Similarly.1 · 10−3 1.11 · 10−3 = Y 1/EN.dc and Ct.91 = −EN. laws (0a) for p = 1 and (4) are inverse relations of each other.dc CN. There is a similar relationship to eq.85 1. (16) where the asterisk stands for {tm.

They provide a general self-consistent rendering of stochastic time series.2%. 5%. 186%. respectively.tck . Considering thresholds of 0.4%. AUD-USD 10 . dc.tck Et. 5% the new average coastline lengths are now 1.1%. (17) where the empirical values lead to Et. respectively. 2. as for example law (9) (tm) could be derived from law (12) (tm). and the relation given in equation (17). (5). (12) (tm. 5%] and measuring it for 0. The range of these average daily coastline lengths is from 1. (9) (tm. it is still possible to evaluate the length of the coastline by employing the scaling relation for the interval [0. 2002 to December 1. 34. The 500-fold decrease in resolution entails a coastline decrease by a factor of only 220. From these. 264%. and specifically give a glimpse of the bare bones of the market structure.96. a tick) every 258 seconds.8%. The following currency pairs are considered with the total number of ticks given in parenthesis: AUD-JPY (15’286’858). 3 3. 65.1%. for the thresholds 0. os). 1′946%. 34.3 The coastline We now have the necessary tools in hand to come back to the measurement of the length of the coastline. This means there is a tick to be expected every 279 seconds.01%. 2007. highlighting the fact that GRW has fewer small moves and more middle-sized moves than the empirical price curves.e. dc. dc.. 1%. (2). taking transaction costs into account breaks the scaling law for small thresholds. looking at the GRW we find 14′ 361%. from December 1.8% for EUR-CHF to 9. 2′ 046%. the length of the coastline decreases by a factor of 650. (10) (tm. and (11) (tm. and Ct. This expectation is compared with law (4) which indicates a move of 0. Similarly. os).1% for AUD-JPY. 2). Thus.tck = ∆t Ct.02% (i.05%. os).1 Methods and data The data set We use a tick-by-tick database composed of 13 currency pairs spanning five years. In summary. (3) (p = 1.05%. Note that our classification is somewhat arbitrary. we measure an average daily move of 6.05% threshold (which occurs on average every 15 minutes). one finds the average lengths of the annualised coastline to be 22′ 509%. However.1%.5%. 0. 1.tck = 0.(0a) and (1) can be used to relate the average number of ticks to a time interval ∆t hN(∆xtck )i = ∆t 1/E (1) Cx (1)CN. plus laws (0a) and (0b). For the 0. So by decreasing the threshold of resolution 500-fold. 161%. 1%. 0. os). The total-move scaling law (12) allows us to estimate its size as a function of the resolution defined by the directional-change threshold.tckx !Ex (1)EN. 463%. six additional ones can be derived: (4). Not surprisingly. There are 12 main independent scaling laws: (1). 604%.tck = 279. we have uncovered 18 novel empirical scaling-law relations.

GBP-JPY (26’423’199). (18) where x0 = 1. xext = x0 . EUR-AUD (19’111’129). USD-CHF (13’812’055). i. Performing this calculation for multiple thresholds recovers law (0b). 3. σ 2). 11 . ∆t = 1 second. Excluding law (3). This. USDJPY (13’507’173). algorithm 2 counts the number of directional changes given a threshold ∆xdc . Similarly. a simple Gaussian random walk (GRW) model consisting of one million ticks is considered as a benchmark ∆xi = x(ti + ∆t) − x(ti ) ∼ N (0.. however. The difference in the number of ticks is due to varying liquidity and the fact that some exchange rates are synthetically generated from two data streams.e. The approximately 26. EUR-GBP (13’847’688).336723. This setting is arbitrarily chosen so as to mimic realistic FX behaviour.6. GBP-CHF (27’141’146). we provide the USD-JPY and EUR-USD datasets that were used in our analysis at www. In addition to the empirical data.2 Pseudocode Algorithm 1 computes the number of prices moves n = n↑ + n↓ for a fixed percentage threshold ∆x. As the timing of price quotes does not usually coincide with the fixed sampling times implied by an interval ∆t. The data is filtered as reoccurring ticks showing the same price as the last registered tick are omitted. and σ = 1/6769.ch/more/datasets/. As an example.olsen.Algorithm 1 priceMoveCount(x) Require: initialise variables (n↑ = n↓ = 0. Applying this computation to an array of thresholds yields law (2). CHF-JPY (17’081’987). GBPJPY is derived from GBP-USD and USD-JPY. the computation of all the other new scaling laws featured in this study rely on the detection of price moves and directional changes. ∆x ≥ 0 fixed) 1: if (x − xext )/xext ≥ ∆x then 2: n↑ ← n↑ + 1 3: xext ← x 4: else if (x − xext )/xext ≤ −∆x then 5: n↓ ← n↓ + 1 6: xext ← x 7: end if (7’037’203). GBP-USD (13’918’523). EUR-USD (13’093’081). EUR-CHF (9’912’921).5 million ticks in the GBP-JPY cross-rate are derived from the ticks in GBP-USD (13.5 million). market makers quote prices based on two feeds creating cross-rates. does not mean that the GBP-JPY data is artificial as it is also real published market data. To facilitate the reproduction of our results.9 million) and USD-JPY (13. for all scaling laws proportional to a power of ∆t we use an interpolation scheme which considers the last quoted price. EUR-JPY (22’594’396).

we report the adjusted R2 values of the fits. For laws depending on time intervals. where A and B are the unknown parameters to be estimated. As seen in the insets of figure 2 it is unclear to what family of distributions they belong. xext = x0 .. i.r-project.3 Data fitting It is worth noting that we do not attempt to fit power-law distributions to empirical data (see Clauset et al. and the range is from 20 to 3′ 975′783 seconds (which is 46 days. Y = A + BX + CX 2 . We select 250 measurement points for laws proportional to price thresholds.e.05% in logarithmic steps: in log-space the difference of the threshold values is always 0. As scaling-law relations are identified from their linear ∗ www.. For some of the plots a slight curvature can be discerned (e. In addition to the linear model. Rather.g.025. to detect systematic curvature in the fitted data.05. mode is up. plus their errors. or Y = A+BX. we detect scaling-law relations for the average and cumulative values of various quantities uncovered in the empirical data.01% to 5. for EUR-CHF in figures 3a and b). 12 . 23 minutes and 3 seconds). In addition. and the range is from 0. The actual fitting is done using the R programming language’s∗ linear model function. We assume a linear relationship between the response variable Y and the random variables X. there are 245 observations. The logarithmic steps are always 0. ∆xdc ≥ 0 fixed) 1: if mode is down then 2: if x < xext then 3: xext ← x 4: else if (x − xext )/xext ≥ ∆x then 5: n↑ ← n↑ + 1 6: xext ← x 7: mode ← up 8: end if 9: else if mode is up then 10: if x > xext then 11: xext ← x 12: else if (x − xext )/xext ≤ −∆x then 13: n↓ ← n↓ + 1 14: xext ← x 15: mode ← down 16: end if 17: end if 3.Algorithm 2 directionalChangeCount(x) Require: initialise variables (n↑ = n↓ = 0. We actually make no claim on how the data is distributed for each predefined point of observation. a quadratic model is tested as an alternative hypothesis. The tables provided in appendix A give the estimated scaling-law parameters for all 13 currency pairs and a GRW model.org. [2009]).

In the last column we check for any curvature by comparing the quadratic model’s R2 value to the 2 2 previously reported linear value. Finally. we systematically test if a quadratic model yields a better fit. To see how the error of C 2 propagates.. it is straightforward to retrieve a scaling law relation: x E . E = B. B ∼ N (µA. (19) y= C where y = exp Y .dependence. the last row shows the average parameter values for the currency pairs and in parentheses the sample standard deviations.B ) and use the approximation !2 !2 . we assume A. The quadratic model yields mostly a marginally improved fit only for the GRW data. and C = exp(−A/B).e.B . Rquad − Rlin . σA. x = exp X. From the linear model. i.

.

∂f .

.

∂f .

.

(20) ∂A . Var[C] ≈ σA + σB .

µB ∂B .µA .

In finance.60% to be observed within 24 hours (law (3). measuring risk. and the Seminar for Statistics at the ETH in Zurich for their consultation. 5 Acknowledgements We thank L.4% to be measured (law (4)). and taking costs into account. Emax (1). where frames of reference and fixed points are hard to come by and often illusory. This indicates the importance of considering not only the tail events associated with crisis. holding for close to three orders of magnitude and across 13 currency pairs. but also accounting for the numerous smaller events that precede this event. Cmax (1)).05% threshold. 13 . 4 Conclusions We have enlarged the catalogue of FX stylised facts by observing 12 independent new scaling laws. Relationships amongst the scaling laws can be derived and combinations of them yield new laws. the coastline measures on average 6. Wilkens for helping with figure 1. The new laws represent the foundation of a completely new generation of tools for studying volatility. to our knowledge. on average across all currency pairs there is a av av mean maximal move of 0. This is astonishingly long. and creating better forecasting and trading models. In contrast. has not been mentioned in the literature. Grizzard for editing the manuscript. instead of focusing on their stochastic nature. B) = exp(−A/B). and on average it takes 220 days for a move of 6. T. Considering an 0. next to six secondary ones.µA . We believe they can enhance our study of the dynamic behaviour of markets and improve the quality of the inferences and predictions we make about the behaviour of prices. the new scaling laws provide a reliable framework.µB where f (A. Our analysis relied heavily on understanding the empirical time series as an event-based process. and.4% per day.

3e-04 4.928 Currency average -2.382e-04 (2.99901 4.904 -1.7e-03 2.440e+00 8.722e-06 1.422e+00 6.99946 0.068e-04 6.178 -2.908 -2.031e-03 4.2e-03 9.04 ∆EN. R2 2 2 Rquad − Rlin 0.775e-04 4.0e-03 3.265e-04 -1.112e-02 2.tck 1.99857 2.99844 0.879e-02 2.099e-02 1.975 Currency average 2.99904 0.A Tables Table A1: Directional change count.004e-03 1.2e-02 8.8e-03 5.035e-02 2.99818 0.920e-02 2.99337 0.150e-02 1.864 1.1e-04 1.7e-04 1.090e-04 1.9e-03 5.99931 0.642e-04 1.002 -1.1e-02 2.99955 0.793e-06 6.4e-03 3.0e-02 2.133 -2.99740 0.928 2.931e-02 1.99906 0.3e-03 4.6e-04 1.99877 0.028e-02 2.3e-03 3.02% Currency EN.406e+00 9. R2 2 2 Rquad − Rlin 0.067 -2.957e-05 -3.158 -2.4e-03 7.5e-03) Adj.056e-04 7.6e-03 4.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± CN.199e-04 4.7e-02 6.1e-02 7.1e-03 3.070e+01 9.61e+00 ∆CN.6e-02) CN.99871 0.027 1.392e-04 1. law (0b) Currency EN.99932 0.0e-03 3.4e-03 3.083e+00 9.8e-04 2.085 2.7e-02 5.99906 0.0e-04 (1.947e+00 1.3e+00) Table A2: Tick count.3e-02 8.997 1.430e+00 9.0e-02 14 ∆CN.120 2.807e-04 1.137e-03 3.0e-04 1.8e-01 6.7e-02 Adj.136e+01 8.tck ± ± ± ± ± ± ± ± ± ± ± ± ± ± 4.99949 0.864e-02 2.2e-03 3.0e-03 2.dc AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY -2. law (1).1e-02 4.3e-02 7.699e-03 1.3e-04 1.528e+01 1.tck AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.9e-03 (8.841e+00 (1.99696 0.2e-03 5.5e-02 4.7e-03 5.797 -1.932 1.017 -1.838e-03 2.945 1.2e-03 9.3e-04 1.949 -2.4e-03 4.663e-02 2.233e+00 3. ∆xtck = 0.540e-05 2.027e-02 2.206e-02 2.053e-02 2.0e-03 3.99949 0.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± 8.99873 0.99833 0.497e-04 .960e-04 1.283e-04 1.6e-03 2.99950 0.99928 0.03 ∆EN.733e-06 2.4e-03 4.99967 0.99827 0.562e-04 5.046 -1.0e-03 3.152e-05 -2.116e+01 1.0e-01) 8.99931 0.99541 0.827e-04 1.134 2.484e-05 1.2e-04 1.99933 0.051 1.tck ± ± ± ± ± ± ± ± ± ± ± ± ± ± 1.185 1.2e-03 6.970 2.8e-03 5.2e-03 3.2e-04 1.1e-02 5.218e+00 5.5e-04 2.879e-04 7.dc 4.699e+00 8.131 -2.1e-04 3.146e-02 1.99848 0.9e-04 2.122 2.6e-03 1.8e-02 2.908 -1.

9e-04 3.2e-03 4.185e-05 1.988e-04 (2.6e+06) 2 Adj.462 0.469e-04 2.033e-04 1.9e-02 Adj.022e+01 9.2e+04 3.6e-02 4.7e-04 9.99913 0.171e+00 5.x 4.973 -2.420e-04 .3e-03 6.329e-04 4.496 0.99872 0.99547 0.0e-03 1.6e-04 9.1e-03 4.1e-03 9.99802 0.483 0.2e-02 7.99907 0.x ± ± ± ± ± ± ± ± ± ± ± ± ± ± CN.779e+05 7.8e+04 2.99809 0.6e-01 5.048e+00 (8.912e+00 3.758e+00 1.362e+06 1.506e-04 8.001 -1.99737 0.99907 0.142e+00 8.637e+05 7. law (2) Currency EN.2e+05 1.521e-04 5.8e-03 7.99953 0.99837 0.3e+04 1.3e-03 1.7e-02) 1.115e-05 1.x AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY -2.99965 0.0e-03 1.540e-05 1.99931 0.8e-02 1.99897 4.5e+04 8.727e-04 3.x ± ± ± ± ± ± ± ± ± ± ± ± ± ± 7.138 -2.606e-04 7.3e-02 6.632e+05 1.070e+04 5.930 -2.8e-04 7.3e-02 4.493 0.451 0.162e-04 -3.108e-03 6.317e+06 6.456 0.907e-03 2.261e-03 2.3e+00) Table A4: Mean price move during ∆t. R2 2 2 Rquad − Rlin 0.5e+04 4.141e+00 9.99875 0.725e-04 1.1e-03 8.095e+00 7.8e-03 5.516e+05 6.1e+05 1.584e+05 6.2e+04 (1. Rlin 2 2 Rquad − Rlin 0.99200 0.462 0.131e-04 3.0e-03 3.1e-02 2.99932 0.482 Currency average 0.533e-05 -2.092 -2.497 0.949e+05 (1.1e-02 5.Table A3: Price move count.8e+04 2.936 -1.419e+01 1.125e+01 8.517e+05 1.9e-02 6.115e-03 2.450 0.4e+02 1.899e+05 6.051 -1.8e-03 2.8e-03 9.8e+04 2.6e-04 1.8e-03 1.3e-03 2.198e-03 1.8e-03 3.0e+04 3.029 -1.99538 0.901e-06 8.99907 0.99907 0.476 0.5e-02 4.638e-03 1.233e-03 3.3e-03 2.825e+06 6.99903 0.510 0.1e-03 3.866 -1.36e+00 ∆CN.99630 0.210e+00 9.416e-04 2.1e+04 5.978 Currency average -2.1e+04 4.99909 1.99936 0.24e+06 15 ∆Cx (1) ± ± ± ± ± ± ± ± ± ± ± ± ± ± 2.3e-02 1.187 -2.469e+00 6.99951 0.137e+00 9.99951 0.962e+05 8.47 ∆Ex (1) ± ± ± ± ± ± ± ± ± ± ± ± ± ± Cx (1) 1.135 -2.99875 0.1e-02 3.461 0.124 -2.1e-03 1.668e-07 9.946 -1.4e-03 4. p = 1 Currency Ex (1) AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 0.111e+01 1.783e+05 4.2e-03 3.99646 0.0e-03 2.802e-04 3. law (0a).04 ∆EN.473 0.3e-04 4.8e-02) 8.99940 0.825e-04 2.6e-02 5.

p = 1 Currency Emax (1) ∆Emax (1) AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 0.616e+05 4.3e+04 1.5e+03 1.99950 0.572e-06 9.2e-04 9.99884 0.4e+03 5.472 0.7e+04 7.4e+03 3.6e-04 6.9e+03 3.99880 0.450 0.1e-03 1.99977 0.511 0.852e+05 3.869e-05 5.410e+05 1.995e-04 1.99867 0.99548 0.7e+03 (2.515e+05 1.467 0.99947 0.99727 0.5e-03 6.743e+04 1.8e-04 1.8e-04 2.508 Currency average 0.99865 5.016e-04 7.654e-06 2.2e-03 9.762e-05 -1. Rlin 2 2 Rquad − Rlin 0.440 0.676e+05 2.042e-06 8.516 0.7e+04 8.463 Currency average 0.46 ∆Ex (2) ± ± ± ± ± ± ± ± ± ± ± ± ± ± Cx (2) 9.454 0.6e+05) .6e-04 7.195e-04 8.3e+03 6. R2 2 2 Rquad − Rlin ± ± ± ± ± ± ± ± ± ± ± ± ± ± 0.473 0.487 0.99936 0.888e+06 1.473e-03 -1.458 0.131e+05 3.3e-03 7.588e+05 4.130e-05 1.018e+06 3.679e-04 6.399e+05 7.225e+05 4.4e+03 1.451e-04 3.976e-04 1.99956 0.467 0.429e-04 3.467e+05 3.99910 0.99939 0.5e-04 2.2e-03 1.495 0.99952 3.274e+05 2.685e-03 1.212e+05 3.6e-04 6.5e-04 1. law (0a).99730 0.084e+06 3.617e-04 8.795e-04 1.355e+05 1.057e-05 8.99852 0.99867 0.572e+05 1.757e-06 4.8e+03 1.1e+04 1.49 ± ± ± ± ± ± ± ± ± ± ± ± ± ± Cmax (1) 1.4e+03 5.0e+04 4.1e-04 1.479 0.1e-03 1.464 0.9e+03 (9.426e+05 1.4e+03 1.469 0.99937 0.106e+05 1.8e+04 2.Table A5: Quadratic mean price move (historical volatility) during ∆t.99942 0.521 0.4e+03 3.99914 0.513 0. p = 2 Currency Ex (2) AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 0.0e+04 1.441 0.528e+05 1.303e-04 8.5e+03 7.99955 0.7e+03 6.500 0.522 0.133e+03 3.830e-04 3.996e+03 1.475 0.1e-03 4.464e-04 -2.2e-04 4.3e+05 5.2e-04 6.4e-02) 7.190e-04 8.1e-03 7.99908 0.2e-03 8.1e+03 8.579e+05 (2.957e-05 7.466 0.888e-05 7.99822 0.1e-03 1.449 0.6e-04 1.8e+03 5.912e+05 (1.99892 0.3e-04 6.412e-06 6.5e-04 1.99951 0.478 0.463 0.1e+03 4.8e+05) Table A6: Maximal price move during ∆t.2e-02) 2.4e-03 1.1e+02 8.843e+05 7.020e-03 3.99893 0.99872 0. law (3).2e-03 7.430 0.873e+05 2.2e+01 4.12e+05 ∆Cx (2) ± ± ± ± ± ± ± ± ± ± ± ± ± ± 2 Adj.410e-06 3.99980 0.41e+05 16 ∆Cmax (1) Adj.523e+05 1.0e-03 1.7e+04 5.3e-04 7.

416e-04 1.1e-05 2.99940 0.627e-03 1.x 4.213e-04 4.864 1.3e+05) Table A8: Time of price move.513e+04 1.99905 0.99954 0.485 Currency average 0.5e-05 1.439e+05 2.584e-06 1.99957 0.050e+05 1.7e-04) .x Adj.99939 0.739e+03 1.337e+05 2. p = 2 Currency Emax (2) ∆Emax (2) AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 0.0e-03 1.0e-05 0.459e-05 2.336e-06 6.99835 0.2e-03 3.028 1.594e-06 3.932 1.125 2.237e-04 -1.0e+03 2.333e-04 2.774e-03 2.99740 0.198e+05 1.845e-04 6.99963 0.99942 0.6e+03 2.1e+03 (2.227e-03 1.843e-03 2.8e-04 4.0e-04 6.267e-04 2.4e+04 4.229e+05 1.398e-04 1.477e-03 1.99959 0.7e-05 9.481 0.98e+05 ∆Cmax (2) Adj.99967 0.99933 0.5e+03 4.7e+03 1.6e-04 7.1e-03 9.598e-04 1.470 0. R2 2 2 Rquad − Rlin 3.618e-04 2.387e-04 8.725e-03 1.475 0.6e+03 6.99871 0.9e-03 3.99907 0.4e-03 4. law (3).99932 0.945 1.4e-04 9.4e-05 1.450 0.5e-04 7.7e-05 1. law (4) Currency Et.495 0.0e+03 2.977 Currency average 2.99887 0.0e-04 6.191e+05 9.839e-04 2.5e-06 1.451e-03 9.99980 0.8e-0) Cmax (2) 6.051 1.705e-05 -3.259e+05 1.1e-03 1.7e-04 1.853e-04 1.2e-03 4.99914 0.088 2.162e-03 8.461e-03 (8.814e-04 (4.494 0.99898 4.134 2.x ± ± ± ± ± ± ± ± ± ± ± ± ± ± Ct.493 0.8e+03 4.6e+03 4.850e+04 8.5e-04 1.145e-04 1.080e+05 1.0e-03 2.721e-05 1.Table A7: Quadratic maximal price move during ∆t.428e-03 1.99951 0.803e-04 6.7e-03 2.1e-05 1.794e-04 1.99947 0.277e-04 1.99907 0.555e-04 1.042e-03 8.1e-03 5.510 0.721e-06 2.999 1.415e-04 1.4e+03 2.7e-05 5.640e-03 1.466 0.0e-03 3.99982 0.7e-04 3.x AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.0e-03 (1.0e-03 2.486 0.8e-03 5.529e+05 2.123 2.4728 ± ± ± ± ± ± ± ± ± ± ± ± ± ± 7.4e-04 1.99858 0.2e-03 5.825e-03 1.184 1.841e-03 1.928 2.71e-03 17 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ∆Ct.99888 2.2e-03 6.99538 0.4e-03 2.450 0.106e+05 1.5e+01 3. R2 2 2 Rquad − Rlin ± ± ± ± ± ± ± ± ± ± ± ± ± ± 0.4e+03 3.6e-05 2.6e-05 3.3e-05 3.972 2.0e-03 3.806e+05 1.905e-03 1.6e-02) 1.9e+03 2.109e-03 2.446 0.5e-04 6.6e-05 1.99906 0.04 ∆Et.99947 0.456 0.

4e-05 2.99947 0.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± Ct.5e-03 1.608e-03 2.tm 2.708e-01 4.4e-03 2.046 1.885e-01 4.1e-03 2.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± 3.438e-04 6.926e-03 1.6e-05 1.2e-03 3.99963 0.4e-03 4.848e-05 3.9e-03 3.790 1.016 1.066e-04 2.99868 0.0e-03 1.081e-04 .99913 4.995 0.652e-04 6.063 2.99930 0.814e-04 2.641e-04 1.5e-05 2.993 0.99899 0.99872 0.6e-03 1.127 2.053e-03 2.99873 0.3e-03 2.299e-01 5.2e-03 9. law (9).99851 0.450e-04 7.725e-05 2.953e-03 1.99857 2.819e-04 1.3e-03 2.899 1.99932 0.111 2.2e-05 4. R2 2 2 Rquad − Rlin 0.1e-03 3.7e-03 1.99956 0.835e-06 1.388e-01 5.969e-05 3.9e-03 3.996 0.5e-03 4.001 0.99741 0.99950 0.99670 0. R2 2 2 Rquad − Rlin 0.749e-04 4.180e-04 (5.8e-03 2.99863 0.981 0.324e-04 4.8e-03 6.180e-04 3.531e-03 9.8e-03 4.375e-05 -2.dc AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.99940 0.266e-03 (9.020e-03 6.99877 0.740e-04 2.1e-05 Adj.097e-05 2.997 1.6e-03 2.Table A9: Time between directional changes.2e-03 1.3e-03 1.9e-02) Adj.2e-04) Table A10: Total price move.2e-03 1.4e-03 2.132 2.876e-01 5.9e-03 1.060e-03 -1.247e-03 1.804e-03 1.99361 0.99893 0.dc 4.211e-01 5.927 Currency average 2.042e-05 1.039e-01 4.02 ∆Et.976 0.99865 0.4e-03 1.358e-01 5.6e-05 1.948 2.2e-03 3.4e-03 1.629e-03 1.5e-05 3.006 1.9e-04 1.695e-06 2.0e-03 2.536e-03 1.969 Currency average 0.643e-04 4.5e-05 2.1e-03 3.99921 0.9e-05 2.99782 0.990 0.884 2.943 0.001 1.586e-03 1.162 2.7e-03 2. law (5) Currency Et.tm AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 1.050e-03 1.439e-03 1.5e-03 4.973 0.904 1.9e-05 1.99928 0.326e-04 7.028e-01 (1.0e-03 3.3e-05 4.3e-03 3. ∗ = tm Currency Ex.232e-01 5.907e-03 1.0e-05 1.2e-02) 5.004 1.570e-06 1.984e-01 5.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± Cx.9e-03 3.258e-04 2.807e-03 2.8e-02) 1.8e-03 8.3e-03 2.99817 0.001e-01 4.4e-03 3.9e-03 1.65e-03 ∆Ct.99932 0.6e-03 2.871e-01 5.10e-01 18 ∆Cx.99635 0.3e-03 (1.99 ∆Ex.001 0.1e-03 4.8e-03 5.998e-01 4.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± 2.99944 0.3e-05 1.

918e-01 9.6e-03 (6.99880 0.072e-04 1.052 0.99922 0.9e-03 2.164e+00 1.098e+00 1.1e-03 1.918e-01 9.064 1.959 0.268e-04 4.2e-03 4.99828 0.dc 2.os ± ± ± ± ± ± ± ± ± ± ± ± ± ± 5.99940 0.926e-01 9.6e-03 1.7e-03 2.4e-03 2.851e-04 3.954 Currency average 0.071 1.99390 0.6e-03) 9.8e-02) Adj.118e+00 1.0e-03 7.6e-03 1.945 0.3e-03 9.847e-01 9.042 1.os 2.89e-01 ∆Cx.Currency Table A11: Directional-change move.121e+00 1.6e-03 Adj.8e-02) 1.8e-02 6.930e-05 2. ∗ = dc Ex.110e-03 1.943e-01 9.921e-01 9.827e-01 9.941e-04 5.6e-03 3.1e-03 1.7e-03 1.99926 0.99825 -4.937 0.871e-01 9.002e-03 3.879e-01 1.99858 1.002e-04 2.4e-03 5.778e-04 1.0e-03 1.0e-03 4.99816 0.99901 0.004e+00 9.os AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 1.99629 0. ∗ = os Currency Ex.996e-04 (4. law (9). R2 2 2 Rquad − Rlin 0.003 0.6e-03 2.6e-03 6.958 0.99902 0.os ± ± ± ± ± ± ± ± ± ± ± ± ± ± Cx.066 1.045e+00 1.4e-03 2.041 1.0e-03 2.012e+00 9.6e-03 1.99805 0.167e-03 9.812e-01 1.814e-01 9.0e-03 4.5e-03 4.060 1.706e-06 3.0e-03 3.991e-04 4.625e-04 .99900 0.1e-02 7.06e+00 19 ∆Cx.473e-04 8.959 0.901e-06 1.99818 0.98905 0. R2 2 2 Rquad − Rlin 0.8e-03 6.1e-03 3.059e-03 1.996 1.3e-03 4.5e-03 2.934e-01 (8.0e-03 6.990 0.6e-03 1.2e-03 4.7e-03 2.267e-04 8.6e-03 5.963e-03 7.7e-03 1.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± 6.1e-02 6.937 0.656e-03 6.9e-03 1.5e-03 2.99937 0.3e-03 1.8e-03 4.898e-01 9.5e-03 6.081 1.949 0.913e-01 9.488e-04 5.946 0.99865 0.8e-03 2.941 0.961 0. law (9).961 0.987 Currency average 1.123e-03 -9.027e+00 9.5e-03 2.3e-03 5.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± Cx.974e-04 6.907e-01 9.718e-04 1.9e-03 3.95 ∆Ex.5e-02 1.5e-03 2.273e-04 3.168e+00 1.941e-04 4.788e-01 1.99823 0.dc AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 0.843e-01 9.103 1.673e-04 8.961 0.067e+00 (3.99692 0.99830 0.2e-03) Table A12: Overshoot move.99838 0.2e-02 9.1e-03 5.6e-03 1.99815 0.948 0.9e-03 2.99798 0.918e-01 9.99228 0.99478 0.7e-03 6.9e-03 6.99884 0.99934 0.033e-03 9.8e-03 4.04 ∆Ex.964 0.

5e-03 4.124 2.099 1.866 1.229e-03 1.004 1.282e-05 2.871 2.063e-03 2.99952 0.dc 4.4e-05 2.4e-03 9.167 2.02 Currency ∆Et.626e-03 1.99931 0.066 2.99885 2.925 Currency average 2.99875 0.5e-05 2.1e-03 3.311e-03 8.135 1.035e-04 1.99960 0.450e-03 1.103 2.5e-05 3.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± Ct.2e-05 5.99240 0.99882 0.967e-03 1.133 2.3e-03 3.5e-03 4.8e-03 2.99786 0.tm 4.596e-03 1.4e-05 2.4e-05 2.Table A13: Time of total move.99816 0.789e-03 1.791 1.616e-03 2.768e-05 5.926e-03 1.9e-05 (7.8e-03 3.448e-04 4.5e-03 4.65e-03 ∆Ct.99857 3.99887 0.813e-05 1.016e-03 2.99941 0.4e-03 3.846 2.439e-03 5.8e-03 4.8e-03 5.0e-05 1.99915 0.1e-05 Adj.99876 0.4e-04 2.018 1.4e-03 3.9e-03 3.958 1.415e-03 1.996 1.3e-03 4.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± 5.809e-03 1.251e-04 2.8e-05 1.053e-03 1.3e-05 5.183e-04 1.5e-03 8.99594 0.1e-05 3.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± Ct.578e-04 6.914 Currency average 1.99812 0.49e-03 20 ∆Ct.651e-06 1.129e-04 5.1e-03 3.6e-05 2.774 1. ∗ = tm Currency Et.430e-04 6.7e-05 1.073e-04 1.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± 3.307e-04 (5.935e-04 .1e-03 2.919e-03 1.023e-03 4.99934 0.280e-03 3.143e-04 6.8e-03 8.638e-03 1.127e-04 5.463e-03 2.0e-04 1. R2 2 2 Rquad − Rlin 0.054e-04 6.1e-05 4.015 2.655e-04 1.838e-03 2.99870 0.870e-03 1.5e-05 5. law (10).1e-03 4.3e-05 2.0e-05 2.636e-03 2.1e-03 6.99938 0.2e-03 9.534e-03 9.6e-03 2.7e-03 7.3e-05 3.950 2. law (10).dc AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 1.0e-05 2.98 ∆Et.654e-03 2.9e-03 3.980e-05 3.tm AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.898 1.066 2.1e-03 5.102 2. ∗ = dc Et.99709 0.888 1.0e-05 3.0e-01) 2.902 1.6e-03 4.886 2.8e-02) 1.705e-04 8.941 1.049 1.99653 0.3e-04) Table A14: Time of directional change.3e-03 3. R2 2 2 Rquad − Rlin 0.99914 0.874e-03 2.9e-05 4.1e-03 2.99868 0.6e-05 1.5e-04) Adj.149e-04 3.99892 0.99875 0.651e-05 -2.99928 0.99338 0.737e-05 6.235e-02 2.026e-03 6.99945 0.057e-03 (1.1e-05 3.4e-05 3.260e-03 (9.089e-03 1.3e-05 2.987e-04 4.096e-03 3.

os ± ± ± ± ± ± ± ± ± ± ± ± ± ± 5.997 2.7e-05 3.0e-02 1.473e-04 5.99810 0.997e-04 1. law (11).411e-03 2.098e-02 2.922 Currency average 2.18e-03 ∆Ct.8e-03 1.9e-04) Table A16: Total-move tick count.5e-03) Adj.5e-04 1.1e-04 1.8e-05 2.094e-04 2.9e-04 1.9e-03 4.99937 0.698e-03 3.3e-04 1.454e-04 1.9e-04 3.866e-02 2.765e-06 2.os ± ± ± ± ± ± ± ± ± ± ± ± ± ± Ct.99858 3.274e-03 1.99796 2.99812 0.121 2.519e-03 2. law (10).135e-05 1.919e-05 2.796 1.899e-03 1.os 5.99927 0.931e-02 2.0e-03 4.99813 0.6e-03 4. ∗ = os Currency Et.767e-02 1.5e-04 2.058e-03 9.166e-02 2.7e-05 6.99778 0.372e-03 8.016 1.047e-02 (1.2e-03 4.624e-03 (1.tm AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.8e-04 2.99931 0.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± CN.99420 0.1e-02 3.99338 0.902 1.1e-03 3.4e-05 4.99750 0.234e-03 2. ∗ = tm EN.Table A15: Time of overshoot.4e-03 5.2e-04 (1.273e-03 2.909e-02 1.505e-05 2.5e-04 1.02 ∆EN.617e-04 4.2e-03 9.97e-02 21 ∆CN.4e-05 2.681e-04 3.99441 0.121 2.241e-03 2.5e-03 3.799e-05 2.1e-05 2.4e-03 4.99816 0.99937 0.9e-04 2.900e-03 1.823e-03 2.2e-04 1.4e-04 2.150 2.063 2.661e-02 2.9e-03 5.314e-04 (6.119 2.435e-03 1.552e-04 1.8e-03 6.792 1.99871 0.131 2.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± 1.4e-04 4.225e-02 1.1e-03 5.2e-05 2.049 1.6e-03 4.os AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2. R2 2 2 Rquad − Rlin 0.182e-03 1.99862 0.6e-03 1.597e-03 1.122e-04 4.7e-03 2.6e-04 2.4e-03 9.99884 0.177 2.2e-05 Adj.99891 0.0e-02) 1.604e-04 5.014e-04 5.864e-02 1.008e-02 1.048 1.337e-06 6.906 2.99884 0.923e-04 3.2e-05 7.99831 0.4e-03 3.190 2.949 2.784e-04 1.469e-03 2.136 2.640e-04 .088e-04 6.003 1.861e-02 2.3e-03 7.4e-03 4. R2 2 2 Rquad − Rlin 0.tm 4.930 Currency average 2.049e-04 4.0e-01) 2.5e-04 1.893 2.093 2.035 1.3e-03 3.04 Currency ∆Et.7e-05 3.079 1.2e-05 4.4e-03 3.8e-03 4.176e-04 -2.99950 0.99869 0.5e-03 5.99687 0.342e-03 4.903 1.99877 0.8e-03 6.1e-05 4.898 1.222e-03 2.921 1.3e-05 3.0e-03 5.024e-02 1.883e-02 1.99295 0.99944 0.99873 0.

181 2.0e-04 7. ∗ = os EN.089 2.4e-04 1.554e-02 2.755e-03 2.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± 4.9e-03 8.182 1.462e-03 -8.809 1.187e-02 3.568e-03 7.772e-02 4.416e-02 2.99471 0.dc AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.7e-04 4.0e-04 2.9e-03 4.58e-02 ∆CN.6e-03 1.1e-01) 2.846e-05 1.99687 0.os ± ± ± ± ± ± ± ± ± ± ± ± ± ± CN.5e-04 2.783 1.465e-02 4.2e-04 3.602e-02 2.0e-03 4.405e-02 (1.002 1.8e-04 (1.8e-04 2.98832 0.759e-04 9.os AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY 2.006 1.8e-03) Adj.6e-04 3.606e-04 9.2e-04 4.985 1.365e-04 2. R2 2 2 Rquad − Rlin 0.3e-04 5.998 1.246e-02 2.2e-04 8.743e-02 2.99873 0.205 2.017 2.4e-03 1.4e-04 2.052 Currency average 2. law (11).7e-04 4.220e-02 2.dc 7.99647 0.535e-03 1.0e-04 3.194e-02 2.3e-03 5.99806 0.137 2.5e-03 7.266e-02 (9.126e-03 1.652e-02 3.306e-02 2.6e-03 9. ∗ = dc EN.198 2.99669 0.8e-03 2.4e-03 5.8e-03 8.99451 0.2e-03 7.548e-02 3.99600 0.99873 0.99678 5.39e-02 22 ∆CN.5e-03 6.99828 0.7e-04 2.99427 7.535e-05 7.dc ± ± ± ± ± ± ± ± ± ± ± ± ± ± CN.375e-02 2.1e-03 4.073e-04 (4. law (11).1e-02 4.460e-04 5.99587 0.9e-03 7.2e-02 6.99442 0.os ± ± ± ± ± ± ± ± ± ± ± ± ± ± 2.5e-03) Table A18: Overshoot tick count.5e-02 9.591e-03 4.277e-02 2.99642 0.99799 0.406e-02 3.6e-04 3.081 1.624e-06 2.101 2.969e-05 4.99884 0.063e-04 3.731e-03 7.884 Currency average 2.991 2.99694 0.6e-03 6.Currency Table A17: Directional-change tick count.436e-02 3.2e-03 1.0e-03 5.239e-02 2.99796 0.128 2.0e-03 6.739e-03 7.4e-03 9.99855 0.868 2.9e-03 2.6e-04 1.7e-04 5.4e-04 3.99583 0.1e-04 Adj.395e-02 2.210e-03 .974e-02 3.2e-04 2.8e-04 6.350e-02 3.99034 0.009 2.140e-02 2.203e-03 1.99658 0.981e-07 1.os 4.99902 0.2e-02) 3.99839 0.782e-02 3.896e-04 5.904 2. R2 2 2 Rquad − Rlin 0.896e-02 3.99786 0.7e-03 4.095 2.978 2.048 1.573e-02 2.323e-04 -4.99088 0.06 Currency ∆EN.0e-03 7.889e-05 4.748e-02 3.303e-04 5.6e-04 5.614e-04 1.0e-03 6.878 1.2e-04 4.871 1.0e-04 7.040 2.01 ∆EN.871e-04 2.305e-06 9.5e-03 6.0e-02 1.0e-03 6.

189 -1.99805 0.tm Adj.734e-03 1.601e+02 2.7e-02) 1.5e-01 ± 1.140 -1.99102 0.tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± (9. ∗ = tm Ecum.2e+00 ± 3.99744 3.5e+00 ± 2.6e-03 2.353e-07 9.8e-03 2.9e-03 1.99838 0.6e-01 ± 1.347e-03 2.6e+00 ± 1.231e-06 1.949e-04 1.783e+01 2.997e-03 -3.979 Currency average -0.99941 0.99916 0.98 Ccum.351e+01 1.854e+02 1.1e-03 1.184 -1.0e-03 8.214e+02 1.145 -1.254e-06 7.99850 0.054e-03 1.580e-03 2.99678 0.605e+02 1.868 -0.0e-02 7.9e-03 2.5e+00 ± 4.6e-03 3.1e+00 ± 3.711e+01 1.99583 0.99796 0.9e-03 2.98509 0.994 -0.312e+01 3.254e+01 5.99874 0.99910 0.695e-04 1.98897 0.0e-03 3.5e-03 3.4e-03 2.956 -0.048 -0. fitted from 0.9e-03 2.2e-03 7.087e-03 1.9e-02) 1.99249 0. law (12).tm ± ± ± ± ± ± ± ± ± ± ± ± ± ± (8. law (12).99366 0.4e-03 4.99701 1.650e+02 0.4e-03 4.358e-05 7.7e-03 3.139e+02 2.4e+01 ± 6.tm AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY -0.009e+02 5.8e-03 2.515e+02 1.076 -1.937 -1.3e+01 ± 4.04 Ccum.1e-01 ± 6.99738 0.307e-04 -7.061e-03 9.tm ∆Ccum.201 -1.99582 0.921e+02 0.059 -0.) 23 .522e-03 (7.4e+00 ± 4.861e+02 1.7e-03 5.8e+00 ± 4.034 -0.51e+02 ± 3.5e+00 ± 2.4e-03 3.943 -0.024 -0.302e-03 1.743e+01 2.99938 0.101e-05 1.903e+02 2.822e-03 1.005 -0.916 -0.6e+00 ± 3.810e-04 5.920e+02 1. ∗ = tm.157e+02 2.154e+02 9.99832 0.949 -0.3e-03 4.99868 0. R2 2 2 Rquad − Rlin 2.9e+00 ± 6.378e+02 2.tm AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY hline Currency average -1.8e+00 ± 3.912 -0.99940 0.4e+00 ± 1.7e+00 ± 2.9e-03 2.878e+02 6.99579 0.1e+00 ± 4.99919 0. R2 2 2 Rquad − Rlin 2.140e-04 2.2e-03 9.524e+02 6.4e-03 6.961 -1.666e+02 1.454e+02 1.2e-03 5.282e-03 (9.848e-03 3.5e-01 ± 1.970 -1.939 -0.96693 0.941 -0.322e-04 3.8e+00 ∆Ecum.7e-03 2.980 -1.3e-03 2.02% was introduced.2% Currency Ecum.5e-03 4.789e-02 3.260e-06 4.284e-02 1.929 -0.3e+00 ± 3.915 -0.tm ∆Ccum.82e+02 ± 6.9e+00 ± 3.588e+02 2.tm Adj.4e+01) Table A20: Cumulative cost-adjusted total move.8e+00 ± 2.8e+00 ∆Ecum.99461 0.Currency Table A19: Cumulative total move (coastline).3e+00 ± 5.652e+02 1.9e+00 ± 1.8e+00 ± 4.003e+01 1.138e-03 3.963 -1.357e-04 1.5e+01) (For the GRW a constant spread of 0.99730 0.

493e-03 1.99213 0.7e-03 8.116e+01 8.3e-01 ± 9.874 -0.99436 0.933e-04 (1.6e-03 4. R2 2 2 Rquad − Rlin 3.3e-03 3.20e+01 ± 3.5e+01) .098 -0.630e+02 1.917 -1.124 -1.954 -1.785e+01 0.486e-03 1.940e+01 6.187e-03 1.5e+00 Ecum.609e+02 2.305e-03 5.121 -1.847e+01 9.803e+02 1.99678 0.7e-03 4.256e-02 5.7e-03 1.010e+02 1.314e-03 (8.99797 0.8e-03 3.8e-03 2.9e-01 ± 1.99731 0.014e+01 0.075e+02 5.044 -0.98147 0.4e+01 ± 1.0e-03 3.dc Adj.99474 0.99901 0.0e-02) 9.232 -1.071 -1.365e-03 5.061e+02 3.155e-04 3.4e+01) Table A22: Cumulative overshoot.805e+01 2.851e+01 7.99117 0.99729 0.99367 0.99775 0.983e+01 2.08 ± ± ± ± ± ± ± ± ± ± ± ± ± ± Currency (3.150e-04 1.1e-01) 6.954 -0.5e-03 4.196e+00 1.547e-03 9.dc ∆Ecum.2e+00 ± 1.178 -1.849e+01 8.6e-03 1.3e+00 ± 2.674e-03 1.7e+00 ± 1.324e-04 4.os Adj.3e-03 4.os ∆Ccum.99590 0.2e-01 ± 1.906 -0.99813 0.99872 0.99678 0.108 -1.4e-01 ± 3.9e-03 2.os ∆Ecum.954 -0.9e-03 2. ∗ = os Ccum.99766 0.2e-03 3.98589 0.99867 0.014 -1.054e+02 7.240 -1.Currency Table A21: Cumulative directional change.5e-01 ± 9.922 -0. law (12).578e-04 1.3e-03 3.953e-03 5.1e+00 ± 1.2e+00 ± 8.5e-03 4.1e-01 ± 4.727e-05 2.dc AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY -1.99597 0.117e+01 9.745e-04 7.374e-04 3.503e+01 1.460e-04 2.365e-03 4. ∗ = dc Ccum.3e+01 ± 3.99380 1.2e-03 6.0e-03 4.2e-03 7.945 Currency average -0.62e+01 ± 1.612e+01 5.968 -0.3e-01 ± 2.978 Currency average -1.99667 0.0e-01 ± 9.978 -0.2e+00 ± 1.143e+02 9.8e+00 ± 3.892 -1.8e-03 6.4e-03 3.277e+02 9.737e-04 3.376e+01 7.99831 0.99852 0.9e+00 ± 1.387e+01 4.654e+01 1.3e+00 ± 9.6e-03 1.dc ∆Ccum.99843 0.0e-01 ± 1.471e-03 1.2e-03 7. R2 2 2 Rquad − Rlin 2.3e-01 ± 1.5e-01 ± 1.99413 0.967e-04 4.4e-03 2.3e+00 Ecum.1e-03 2.7e-03 3.308e-05 1.061 -0.os AUD-JPY AUD-USD CHF-JPY EUR-AUD EUR-CHF EUR-GBP EUR-JPY EUR-USD GBP-CHF GBP-JPY GBP-USD GRW USD-CHF USD-JPY -0.399e-03 1.99811 2.117e+01 2.0e-01 ± 7.948 -0.683e-03 3.006 -1.99921 0.3e-03 6.018e-04 2.0e-03 3.99 ± ± ± ± ± ± ± ± ± ± ± ± ± ± 24 (5.084 -0.191 -1.1e-01 ± 9.385e-02 1.6e+00 ± 4. law (12).866 -0.751e+00 2.888e+01 7.2e+00 ± 5.

Financ. 423(6937):267 – 270. J. 2009. R. Multi-scaling in finance. Gopikrishnan. 2007. A. J. Apparent scaling. Lillo. 1(1):105 – 112. M. A.. B.-L. Efficient capital markets: A review of theory and empirical work. M. Ballocchi.-P. M.. 286 (5439):509 – 512. Power laws in economics and finance: some ideas from physics. M. Dacorogna. S. SIAM review. Triennial central bank survey of foreign exchange and derivatives market activity. Quant. O. U. S. M. Financ. 25 . and Y.References G. Dacorogna.. Di Matteo. Peinke. Bouchaud. X. Power-law distributions in empirical data. B.. Newman. 1999. Financ. and R. Bank. 2001. Stoch.. E. 6:479 – 513. 245:423– 436.-C.E. and Y. and O. Dodge. Turbulent cascades in foreign exchange markets. 2004. and H. Quant. F. G. An introduction to high-frequency finance. Gen¸cay. M. Financ. Marsili. Long term memories of developed and emerging markets: using the scaling analysis to characterize their stage of development. Albert. Scaling in currency exchange. 1999. E. Empir. 2001. A. Science. A. 51(4):661–703. Barab´asi and R. Economic Notes – Review of Banking. The intraday multivariate structure of the eurofutures markets. A.. 2005. 381:767 – 770. Quant. 2001. 1997. J. F. Galluccio.. Dacorogna. E. Financ. 4(1):C7 – C11. J. Gabaix. M¨ uller. Olsen. 5(1):103 – 113. San Diego. 2001. Dacorogna. 29(4):827 – 851. V. C. Plerou. Financ. Academic Press. A theory of power-law distributions in financial market fluctuations. and M. Nature. Di Matteo. P. and M. Stanley. Financ. Doyne Farmer and F. Olsen. M. R. J. On the origin of power-law tails in price fluctuations. Hopman. 30(2):183 – 204. P. U. Barndorff-Nielsen and K. M¨ uller. Breymann. R. Ghashghaie. Pictet. Consistent high-precision volatility from high-frequency data. V. Talkner W. Finance and Monetary Economics. T. 2003. Aste. G. J. 2007. T. Caldarelli. Zumbach. Shalizi. U. T. C. 1970. Corsi. Physica A. and M. pages 383–417. Emergence of scaling in random networks. M. 7(1):21 – 36. Fama. Prause. J. Bank for International Settlements. 1996. Nature. M¨ uller. Zhang. Clauset.

Global financial stability report. J.html. West. 2000. B. Contemp. Fokker-planck equation of distributions of financial returns and power laws. Why stock markets crash: critical events in complex financial systems.org/p/zbw/cauewp/5159. D. T. V. Pictet. Mandelbrot and R. 1990. E. Newman. J. A. Morgenegg. 276(5309):122 – 126. The (mis)behavior of markets. Mandelbrot. Financ. O.H. Statistical study of foreign exchange rates. From the bird’s eye to the microscope: A survey of new stylized facts of the intra-daily foreign exchange markets. Guillaume. Grunberg. M. 26 . D. B. Physica A. U. empirical evidence of a price change scaling law. and J. 2008. Lux. Basic Books. L. Power laws.J. 36:394 – 419. M¨ uller. Stoch. M. D. B. Financial power laws: empirical evidence. Economics working papers. models. Enquist.arxiv. Princeton. E. Dacorogna. M. Phys. U. R. 2009. Olsen. 2004. N. B. Bouchaud. Financ. M. Wilmott Magazine. URL www. Bus. M. J. M.D. M¨ uller. 46:1 – 6. The variation of certain speculative prices. B. D.1769. A. Sornette. 2002. R. Stock price jumps: news and volume play a minor role. and O. International Monetary Fund. A. Joulin. 1997. and intraday analysis. 2006. Mantegna and H. R. Scaling behavior in the dynamics of an economic index. A general model for the origin of allometric scaling laws in biology. Science. 1:95 – 129. Olsen. Sornette. Princeton University Press. V. Hudson. Dav´e. Schwarz. Bank. M. A. B.org/abs/0803. Nature. 1995. 46 (5):323 – 351. Stanley. R. 1963.. pareto distributions and zipf’s law. New York.B. Dacorogna. and mechanisms. 376:46 – 49.. URL ideas. Brown. G. 2005. Pictet.repec. J. Lefevre. 14:1189–1208. and C. and B. 1997...-P. 290(1):211 – 217.

- 79315_Lecture 6 - Qualitative and Quantitative SamplingUploaded byFee Syafickham
- Impact of Random Failures and Attacks on Poisson and Power-Law Random NetworksUploaded byComplexNetworks
- BESO2D Command Line UsageUploaded byeduard2891
- PARAMETRICS AND OPTIMIZATION USING ANSOFT HFSS_10.1.1.169.1598[1]Uploaded bydhfsi7496
- BI Integrated Planning - ScenarioUploaded byOmar Amjad
- RF Power Control and Handover Algorithm_ Variable Averaging Window SizeUploaded byBernard P
- WCDMA Prfile 16Uploaded byAnkur Khanna
- Cap05-02Uploaded byJohanAcevedo
- Fianncial MarketsUploaded bybourne
- Basic CreepUploaded byWladimir Gonzalez Velez
- 108.pdfUploaded bynicu1212
- Pareto Law in a Kinetic Model of Market With Random Saving PropensityUploaded byDaniel Quicazán
- 40 Multiple Choice Questions in Basic StatisticsUploaded by29_ramesh170
- Bridge ClassUploaded bysajjettablu
- ST06Uploaded bydillipparida10
- Exam 050272RR Probability a Breeder Records Probabilities for Two Variables in a Population AnswerUploaded byMike Russell
- BA201 Engineering Mathematic UNIT2 - Measures of Central TendencyUploaded byAh Tiang
- A New Curve for Fire Temperature in Compartiment FireUploaded bylordoffire
- Add maths questionsUploaded bydrajshah
- Self Organized Criticality and Phase Locking in Models of Production and Inventory DynamicsUploaded byAndrea Bottasso
- Chap006 NewUploaded byadam
- Acoustical Monitoring of Fish Density, Behavior, And Growth Rate in a Tank(1)Uploaded byArief Nur
- Tutorial 04Uploaded byTaybah Foedoerovna
- 100 Prisoners LightbulbUploaded bybasilbabu
- Test for NormalityUploaded bycooole
- D800010X062Uploaded byZakaria Ahmed Khan
- Axiomatic o EjemploUploaded byedvillal
- Bayesian Analysis SampleUploaded byTanner Phillips
- BMS MAROSUploaded bySumit Rao
- Creating Product and Process Control SystemsUploaded byNeydi Dávila