1: \begin{abstract}

2:  We study the time change of the relation between the rank 

3:  of a racehorse in the Japan Racing Association and the result of victory 

4:  or defeat. Horses are ranked according to the win bet fractions. As the

5:  vote 

6:  progresses, the racehorses are mixed on the win bet fraction axis.

7:  We see the emergence of a scale invariant relation  between the cumulative 

8:  distribution function of the winning horse $x_{1}$ and that  of 

9:  the losing horse $x_{0}$. $x_{1}\propto x_{0}^{\alpha}$ holds 

10:  in the small win bet fraction region. We also see the efficiency of the 

11:  market as the vote proceeds. However, the convergence to the efficient state 

12:  is not monotonic. The time change of the distribution of a vote 

13: is complicated.

14:  Votes resume concentration on popular horses, after the distribution

15:  spreads to a certain extent. 

16:  In order to explain scale invariance, we introduce a simple 

17:  voting model. In a `double' scaling  limit, we show that the exact scale 

18:  invariance relation $x_{1}=x_{0}^{\alpha}$ holds over the entire range 

19:  $0\le x_{0},x_{1}\le 1$.

20: \end{abstract}

21: