{
  "id": "girsanov-theorem",
  "title": "Girsanov Theorem",
  "lang": "en",
  "source_lang": "en",
  "summary": "Girsanov’s theorem describes how changing a probability measure alters the drift of a stochastic process while preserving its continuous noise structure.",
  "category": "mathematics",
  "subcategory": "probability-statistics",
  "type": "concept",
  "aliases": [
    "Girsanov’s theorem",
    "Cameron–Martin–Girsanov theorem"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/girsanov-theorem/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/girsanov-theorem/",
      "json": "https://aiwiki.page/zh/girsanov-theorem.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The Girsanov theorem is a result in stochastic process theory that describes how a process’s martingale and drift components change when its underlying probability measure is changed. In its best-known form, it states that a Brownian motion with an appropriate added drift becomes a Brownian motion under a suitably reweighted measure. The theorem is fundamental to stochastic differential equations, continuous-time statistical inference, and mathematical finance because it allows different drift models to be related through explicit likelihood ratios."
    },
    {
      "heading": "Changing measure",
      "text": "A change of probability measure assigns new probabilities to events without necessarily changing the underlying sample paths. On a probability space \\((\\Omega,\\mathcal F,P)\\), suppose a measure \\(Q\\) is absolutely continuous with respect to \\(P\\). Its Radon–Nikodym derivative \\(Z=dQ/dP\\) satisfies\n\n\\[Q(A)=E_P[Z\\mathbf 1_A], \\qquad A\\in\\mathcal F,\\]\n\nwhere \\(Z\\geq 0\\) and \\(E_P[Z]=1\\). Thus, for an integrable random variable \\(X\\),\n\n\\[E_Q[X]=E_P[ZX].\\]\n\nWhen information arrives over time through a filtration \\((\\mathcal F_t)\\), the corresponding density process is\n\n\\[Z_t=E_P[Z\\mid\\mathcal F_t].\\]\n\nIt is a nonnegative martingale under \\(P\\).\n\nGirsanov’s theorem identifies the correction needed to turn a \\(P\\)-local martingale into a \\(Q\\)-local martingale. For Brownian motion, this correction is a time integral, so changing measure changes the drift.\n\nIf \\(Z>0\\) almost surely, \\(P\\) and \\(Q\\) are equivalent: they have the same null events. This is stronger than one-way absolute continuity."
    },
    {
      "heading": "Brownian formulation",
      "text": "Fix a finite time horizon \\(T\\). Let \\(W=(W_t)_{0\\leq t\\leq T}\\) be a standard \\(d\\)-dimensional Brownian motion on a filtered probability space satisfying the usual completeness and right-continuity conditions.\n\nLet \\(\\theta_t\\) be an \\(\\mathbb R^d\\)-valued progressively measurable process such that\n\n\\[\\int_0^T\\|\\theta_s\\|^2\\,ds<\\infty \\quad P\\text{-almost surely}.\\]\n\nDefine the stochastic exponential\n\n\\[Z_t= \\exp\\left( -\\int_0^t\\theta_s^\\top\\,dW_s -\\frac12\\int_0^t\\|\\theta_s\\|^2\\,ds \\right).\\]\n\nThe stochastic integral is interpreted in the sense of Itô calculus.\n\nThe process \\(Z\\) is a strictly positive local martingale. Assume additionally that it is a true martingale on \\([0,T]\\), so that \\(E_P[Z_T]=1\\). Define\n\n\\[\\frac{dQ}{dP}\\bigg|_{\\mathcal F_T}=Z_T.\\]\n\nThen\n\n\\[W_t^Q=W_t+\\int_0^t\\theta_s\\,ds\\]\n\nis a standard \\(d\\)-dimensional Brownian motion under \\(Q\\), relative to the same filtration.\n\nThe signs are linked: a negative stochastic integral in the exponential produces a positive drift correction in \\(W^Q\\). An equivalent convention uses a positive stochastic integral and subtracts the drift instead.\n\nWhat is preserved\n\nThe paths of \\(W\\) are not replaced by different paths when the measure changes. Rather, their probabilities are reweighted. The corrected process \\(W^Q\\) has independent, normally distributed increments under \\(Q\\), with\n\n\\[W_t^Q-W_s^Q\\sim N(0,(t-s)I_d).\\]\n\nIts quadratic variation remains\n\n\\[[W^{Q,i},W^{Q,j}]_t=\\delta_{ij}t.\\]\n\nThe added drift has finite variation and therefore does not alter this pathwise quantity."
    },
    {
      "heading": "Why integrability matters",
      "text": "The requirement that \\(Z\\) be a true martingale is essential. A nonnegative local martingale is a supermartingale, and it may satisfy\n\n\\[E_P[Z_T]<1.\\]\n\nIn that case, \\(Z_T\\,dP\\) does not define a probability measure without further modification.\n\nA widely used sufficient condition is Novikov’s condition:\n\n\\[E_P\\left[ \\exp\\left( \\frac12\\int_0^T\\|\\theta_s\\|^2\\,ds \\right) \\right]<\\infty.\\]\n\nThis guarantees that the stochastic exponential is a martingale on the specified finite horizon. In particular, bounded \\(\\theta\\) satisfies this condition.\n\nNovikov’s condition is sufficient, not necessary. Its failure does not by itself invalidate a proposed measure change; it means that another argument is needed to establish the martingale property.\n\nConditions must also be matched to the time horizon. Densities may define equivalent measures on every finite-time sigma-algebra without producing equivalent measures on the sigma-algebra describing the entire infinite-time path."
    },
    {
      "heading": "Effect on stochastic differential equations",
      "text": "Suppose an \\(\\mathbb R^n\\)-valued process satisfies\n\n\\[dX_t=b_t\\,dt+\\sigma_t\\,dW_t,\\]\n\nwhere \\(\\sigma_t\\) is an \\(n\\times d\\) matrix. Under the measure change above,\n\n\\[dW_t=dW_t^Q-\\theta_t\\,dt,\\]\n\nso the same process satisfies\n\n\\[dX_t=(b_t-\\sigma_t\\theta_t)\\,dt+\\sigma_t\\,dW_t^Q.\\]\n\nThus Girsanov’s theorem changes the drift from \\(b_t\\) to \\(b_t-\\sigma_t\\theta_t\\), while leaving the diffusion coefficient unchanged.\n\nTo obtain a target drift \\(\\widetilde b_t\\), one seeks a process \\(\\theta_t\\) satisfying\n\n\\[\\sigma_t\\theta_t=b_t-\\widetilde b_t.\\]\n\nWhen \\(\\sigma_t\\) is square and invertible,\n\n\\[\\theta_t=\\sigma_t^{-1}(b_t-\\widetilde b_t).\\]\n\nIf \\(\\sigma_t\\) is degenerate, not every drift change is possible through an equivalent change of measure. The required drift difference must lie in the image of \\(\\sigma_t\\), and the resulting density must still satisfy the necessary integrability conditions.\n\nConstant-drift example\n\nLet \\(W\\) be a one-dimensional Brownian motion under \\(P\\), and let \\(\\mu\\) be constant. Set\n\n\\[Z_T=\\exp\\left(\\mu W_T-\\frac12\\mu^2T\\right), \\qquad dQ=Z_T\\,dP.\\]\n\nUnder \\(Q\\),\n\n\\[B_t=W_t-\\mu t\\]\n\nis Brownian motion. Consequently,\n\n\\[W_t=B_t+\\mu t\\]\n\nhas drift \\(\\mu\\) under \\(Q\\).\n\nFor a fixed time \\(t\\), \\(W_t\\) therefore has the normal distribution \\(N(\\mu t,t)\\) under \\(Q\\), rather than \\(N(0,t)\\). The theorem describes this change for the whole path up to \\(T\\), not merely for one marginal distribution."
    },
    {
      "heading": "General continuous-martingale form",
      "text": "A broader version uses a continuous local martingale \\(L\\), with \\(L_0=0\\), and its stochastic exponential\n\n\\[\\mathcal E(L)_t = \\exp\\left(L_t-\\frac12\\langle L\\rangle_t\\right).\\]\n\nSuppose \\(Z=\\mathcal E(L)\\) is a true martingale and defines \\(Q\\) on a finite horizon. If \\(M\\) is a continuous \\(P\\)-local martingale, then\n\n\\[M_t-\\langle M,L\\rangle_t\\]\n\nis a continuous \\(Q\\)-local martingale. Here \\(\\langle M,L\\rangle\\) denotes quadratic covariation for continuous local martingales.\n\nThe Brownian statement follows by taking\n\n\\[L_t=-\\int_0^t\\theta_s^\\top\\,dW_s.\\]\n\nThe correction becomes\n\n\\[W_t-\\langle W,L\\rangle_t = W_t+\\int_0^t\\theta_s\\,ds.\\]\n\nMore general versions apply to semimartingales with jumps. There, a change of measure may alter jump intensities or compensators as well as continuous drift. The description “only the drift changes” therefore applies directly to the continuous diffusion setting, not to every version of the theorem."
    },
    {
      "heading": "Proof mechanism",
      "text": "The Brownian theorem combines a martingale transformation with a characterization of Brownian motion.\n\nFirst, the exponential density satisfies\n\n\\[dZ_t=-Z_t\\theta_t^\\top\\,dW_t.\\]\n\nApplying the stochastic product rule to\n\n\\[Z_t\\left(W_t+\\int_0^t\\theta_s\\,ds\\right)\\]\n\nshows that the finite-variation terms cancel. After suitable localization, the change-of-measure identity for conditional expectation establishes that \\(W^Q\\) is a continuous local martingale under \\(Q\\).\n\nSecond, adding a continuous finite-variation process leaves quadratic covariation unchanged:\n\n\\[[W^{Q,i},W^{Q,j}]_t=\\delta_{ij}t.\\]\n\nLévy’s characterization then identifies \\(W^Q\\) as Brownian motion.\n\nThis explains why the theorem requires more than computing a new mean and variance: Brownian motion is characterized by its time-dependent martingale structure and quadratic variation."
    },
    {
      "heading": "Applications",
      "text": "Mathematical finance\n\nGirsanov’s theorem is a principal tool for constructing a risk-neutral measure. For example, suppose\n\n\\[\\frac{dS_t}{S_t}=\\mu_t\\,dt+\\sigma_t\\,dW_t,\\]\n\nand a bank account grows at rate \\(r_t\\). When \\(\\sigma_t\\neq0\\), choosing\n\n\\[\\theta_t=\\frac{\\mu_t-r_t}{\\sigma_t}\\]\n\ngives\n\n\\[\\frac{dS_t}{S_t}=r_t\\,dt+\\sigma_t\\,dW_t^Q,\\]\n\nprovided the density defines an equivalent probability measure.\n\nThe discounted asset is then a local martingale under \\(Q\\). Stronger integrability assumptions may be needed for it to be a true martingale. The construction changes the probability measure used for valuation; it does not assert that the asset’s physical expected return is the bank-account rate.\n\nStatistical inference for diffusions\n\nFor continuously observed diffusion models with a common diffusion coefficient and different drift parameters, Girsanov’s theorem can provide a pathwise likelihood function.\n\nFor example, compare\n\n\\[dX_t=b_0(t,X_t)\\,dt+\\sigma(t,X_t)\\,dW_t\\]\n\nwith a target drift \\(b_1\\). If\n\n\\[b_1-b_0=\\sigma u\\]\n\nand the exponential is a valid density, the likelihood ratio on the underlying filtered space is\n\n\\[\\exp\\left( \\int_0^T u_t^\\top\\,dW_t -\\frac12\\int_0^T\\|u_t\\|^2\\,dt \\right).\\]\n\nWhen this density is measurable with respect to the observed trajectory, it is also the likelihood ratio for the path observation. Such formulas support maximum likelihood estimation for drift parameters.\n\nSimulation and importance sampling\n\nIn importance sampling, a simulation may deliberately change a diffusion’s drift to make rare events more frequent. Girsanov weights then correct expectations back to the original measure.\n\nFor a suitable path functional \\(F\\), equivalence gives\n\n\\[E_P[F]=E_Q\\left[F\\frac{dP}{dQ}\\right].\\]\n\nThis identity underlies drift-based variance reduction in Monte Carlo methods. A valid measure change alone does not guarantee an efficient estimator: poorly chosen weights can have very large or infinite variance.\n\nWeak solutions of stochastic differential equations\n\nThe theorem can also be used to construct a weak solution of a stochastic differential equation by beginning with a simpler process and changing measure to introduce a desired drift. This requires a valid density and does not automatically establish strong existence or pathwise uniqueness."
    },
    {
      "heading": "Historical context and scope",
      "text": "The theorem is named after Igor Girsanov, whose 1960 work established an influential general change-of-measure result for stochastic processes. It is closely related to the earlier Cameron–Martin theorem, which describes how Wiener measure changes under suitable deterministic shifts. Girsanov’s theorem accommodates adapted random drift changes and is naturally formulated using martingales.\n\nIts scope is constrained by the geometry of path space. Equivalent measures preserve almost-sure path properties, including quadratic variation. Consequently, changing drift may produce equivalent diffusion laws, whereas changing a constant nonzero volatility to a different positive value produces mutually singular laws on a continuously observed path space: quadratic variation distinguishes them exactly."
    }
  ],
  "links": [
    "stochastic-process",
    "stochastic-differential-equation",
    "probability-space",
    "random-variable",
    "filtration-probability",
    "martingale",
    "absolute-continuity",
    "wiener-process",
    "stochastic-calculus",
    "quadratic-variation",
    "normal-distribution",
    "conditional-expectation",
    "variance",
    "likelihood-function",
    "maximum-likelihood-estimation",
    "importance-sampling",
    "monte-carlo-method",
    "weak-solution"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}