Modern financial market models typically describe price dynamics over an elementary time interval as the combined effect of two components: a deterministic instantaneous drift and a stochastic (random) increment. The deterministic component reflects risk compensation and behavioral effects such as herding and crowd psychology among traders. The stochastic component represents the noise in price movement, whose amplitude is known as volatility. Volatility itself may be systematic—driven by imitation, sentiment shifts, or other macro-level influences. If the deterministic drift vanishes and volatility remains constant, the stochastic term alone generates random walk trajectories.
Introducing nonlinear dependencies—where both volatility and instantaneous drift depend on past volatility and returns—yields a vast array of possible price evolution paths. Of particular interest are mechanisms generating nonlinear positive feedback loops, where price changes reinforce themselves. Such feedback loops play a critical role in diagnosing price dislocations in currency time series.
Elementary Markov Process
We begin with the elementary Markov process of price formation—but first formalize key definitions.
- Finite stochastic process with independent values: A process where, for any statement p depending only on outcomes up to time ,
(1) P{f = sj | p} = P{f = sj}.
For such processes, past observations do not affect forecasts of future outcomes. For Markov processes, this requirement is relaxed: only the most recent outcome may influence the next prediction. - Finite Markov process: A finite stochastic process satisfying
(2) P{f = sj | (fn−1 = si) ∧ p} = P{f = sj | fn−1 = si}.
(assuming compatibility between fn−1 = si and p). Condition (2) is called the Markov property. Sometimes “Markov process” refers to a continuous-time stochastic process satisfying this property, while definition (2) describes a finite-state Markov chain. - Finite Markov chain: A finite Markov process whose transition probabilities pij(n) are time-invariant (i.e., independent of ).
Transition probabilities at step , denoted pij(n), are defined as:
pij(n) = P[f = sj | fn−1 = si].
A Markov chain can be visualized as a system moving from state to state. It starts in state si with initial probability pij(0). If it occupies state si at any moment, it transitions to sj at the next step with probability pij. Initial probabilities represent likelihoods of each possible starting state.
The initial probability vector π0 = {pj(0)} = {P[f0 = sj]} together with the transition matrix P (with elements pij) fully define the Markov chain as a stochastic process—sufficient to construct the full probability measure over its outcome tree.
The transition matrix of a Markov chain is the matrix P with entries pij. The initial probability vector (or initial distribution) is π0.
Thus, given any probability vector π0 and stochastic matrix P, there exists a unique Markov chain (up to state relabeling) with π0 as its initial distribution and P as its transition matrix.
Illustrative Examples of Markov Processes
To clarify these concepts, consider two practical examples.
1. Market Regime Diagnosis. Suppose the market cannot be bullish for two consecutive days. If the current regime is bullish, the next day is equally likely to be either flat or bearish. If the current regime is flat—or bearish—the next day remains unchanged or switches with equal probability; if it switches, it becomes bullish in half of those cases.
Under these rules, the market is modeled as a three-state Markov chain: S = {Flat (F), Bullish (B), Bearish (M)}, with transition matrix:
2. Finite Asset Price Walk. A currency pair randomly walks within a price corridor bounded by support (S) and resistance (R). Its initial level lies midway between S and R. The price tends upward—toward resistance. Denote the integer distance in pips from S to R as 0 to ; the initial price corresponds to integer i, where 0 < i < .
At each step, the market moves upward (toward state ) with probability p, or downward (toward state 0) with probability q = 1 − p. States 0 and are absorbing. Assuming 0 < p < 1, the transition matrix is:
We now turn to a real-world application on the foreign exchange market.
Markov Chains Triggered by U.S. Nonfarm Payrolls Data
Since initial probabilities reflect the likelihood of each possible “starting state,” an elementary Markov pricing process (or elementary Markov chain) can be interpreted as the market’s instantaneous reaction to an exogenous shock—such as a high-impact news release—with price direction determined by the differential between the actual figure and consensus expectations.
For example, on April 1, 2005, the U.S. Nonfarm Payrolls (NFP) report was released. March’s headline NFP number was +110,000 jobs—well below the forecast of +225,000 and prior reading of +262,000.
Fig. 1. EUR/USD 2-minute chart showing market reaction to the U.S. NFP release on April 1, 2005.
Figure 1 shows the natural EUR/USD response to this “weak” labor data. Market makers reacted to two numbers: the actual print (released at 16:30 MSK) and the expectation already priced in. Since the market had pre-priced +225,000 jobs, the actual +110,000 signaled USD overvaluation. Market makers responded immediately—lifting EUR/USD more than 80 pips within one minute. Subsequently, they encountered aggressive sell orders from bears, halting the rally. Over the next 20–30 minutes, even minor EUR/USD advances triggered renewed bearish attacks—suggesting strong, structurally aligned bearish sentiment among EUR/USD participants.
Fig. 2. Historical distribution of USD/CHF price jumps (past 10 years) at U.S. NFP release (1-, 2-, and 3-minute charts), plus 5- and 20-minute lags.
Figure 2 displays the 10-year frequency distribution of USD/CHF price jumps around U.S. NFP releases—at 1-, 2-, and 3-minute intervals, and at 5- and 20-minute lags after the 16:30 MSK release. At the instant of the announcement, the jump-size distribution is highly non-Gaussian. Within just five minutes, however, frequency sharply concentrates near zero. By 20 minutes, the distribution closely resembles a Gaussian curve. Real-time visualization of this statistical deformation—during the emergence and decay of Markov chains—allows estimation of their duration: on this market, the post-NFP Markov process decays fully within ~25–30 minutes.
Of course, few traders can program real-time fractal statistics measurement—due to software complexity and hardware demands (e.g., massive RAM). A practical alternative is monitoring the Bollinger Bands trend indicator.
A more granular analysis of jump-size frequency—around the 16:30 MSK NFP release window (see Fig. 2, leftmost curves)—across multiple timeframes also helps forecast clustering effects during Markov chain formation. This proves valuable when analyzing the subsequent, post-Markov price process—where memory re-emerges (a long-memory process).
Once Markov pricing processes conclude, market memory returns—and subsequent price action is largely driven by collective trader psychology. When that psychology is strongly structured, robust intermediate trends may emerge, accompanied by price turbulence. Such trends can be effectively modeled using jump-diffusion frameworks—a topic for a separate publication.
FAQ
What is a Markov process in Forex trading?
A Markov process models price movements where the next state depends only on the current state—not the full history—making it ideal for capturing short-term, news-driven market reactions.
How does NFP data trigger a Markov chain in USD/CHF?
The NFP release creates an instantaneous regime shift: price jumps reflect the surprise (actual vs. expected), then evolves through probabilistic transitions until market memory reasserts itself—typically within 25–30 minutes.
Can traders use Markov chains without coding?
Yes—by observing Bollinger Band behavior, clustering of price jumps, and reaction symmetry around high-impact news, traders can infer Markov-like dynamics and anticipate post-news trend persistence or reversal.
