We show that the consideration of algebraic properties of exponential shift families allows one to simplify and unify the procedures of determination of the distribution of a sufficient statistic. We also present a new method for the estimation of a density function based on the determination of the distribution of a sufficient statistic.This paper extends the results of the research reported at the XV Seminar on Stability Problems for Stochastic Models in Perm which were published in [1,2]. Some new results were discussed at the International Conference in honor of N. G. Chebotarev [3]. It was shown in those papers that by the consideration of algebraic properties of exponential shift families one can considerably simplify and unify the procedures of determination of the distribtition of a sufficient statistic in these families as well as solutions of estimation problems. In this paper, an account of these results is presented and one new method for the estimation of a density function is described which does not require any additional statistical procedures besides those of determining the distribution of a sufficient statistic.Let X be a homogeneous space of a coherent Lee group G satisfying the second countability axiom, and let /~ be a quasiinvariant measure with the modular function A on the q-field Bx of Borel sets of A'. Throughout the paper, we shall assume that G acts on ,t' by left shifts and write g-x,, instead of (g. x 1,..., g-z,,) to emphasize that g is a translation of the vector x,~. Definition 1. A family of probability density functions (p.d.f's)on (X, Bx,p) is called a shift family with the generating p.d.f, f(x) and the family ~ = g~0 with ~0 being a random element having f(z) as its p.d.f, will be called a shift family.As a parametric space of a shift family one can take the group G itself, but this is not the best way to introduce parametrization since the correspondence between the elements g E G and the densities f(. I g) = f(g-t')A(g) is not one-to-one. To improve the situation, it is better to consider as a parametric space the set of left co-sets of G with respect to a stationary subgroup of the generating density K = {g: f(z I g) = f(z) for all x E X} (see [5]). This space and its element will be denoted by O and 0, respectively. It is clear that f(. ]gx) = f(" 192) for all gx, g2 E 0, where 0 is a fixed element of e = G/K while the random elements gx~0 and g2~0 may be different. On the contrary, from the analytical point of view it is much more convenient to operate with elements of the group G, so in what follows we shall use the notation f(z I g), g E G, remembering where necessary that f(z ] g) is one and the same for all g E 0, 0 E O. For example, if T,,(Xn) = Tn(X1,...,Xn) is a sufficient statistic for the parameter g E G, then the same statistic is also sufficient for the parameter 0 E O. Moreover, the converse is also true by virtue of the Halmos-Savage characterization theorem (see [6]).Denote by q(t, 9) the main factor in the characterization criterion, namely, letWe shall a...