Peut-on travailler sur les données "approximées" en histoire financière ?

During the nineteenth century, newspapers publish, day after day, the quotation of the government bond, then in the second half of the century, some stocks ; at each year end, one takes the habit to indicate to the investor the highest quotation and the lowest for each month of the last year. These observations have the great advantage to be easily accessible but present nevertheless some serious disadvantages.Having these series of the highest and the lowest quotation, with doubtful precision, as well as the exact observations, we have questionned the usefulness of a careful data counting. It appeared that this gigantic work, in spite of its rigor, brought us relatively minor corrections and that the conclusions rejoined those previously established. That would lead us to conclude that with a light procedure, one could, without real loss in precision, avoid a particularly heavy work of counting.The demonstration, of empirical order, is undertaken in two steps. The first one makes use of classical econometrics: simple regression, Granger-causality and analysis of the seasonality. The second step uses different techniques, the cointegration and the error correction model.