This paper estimates the propensity to retire and the persistence of remaining retired once the decision to retire has been made in the US labour market, using a dynamic panel probit model. The estimated income effect of higher housing wealth is virtually zero and that of financial assets wealth is positive, increasing the retirement probability. The substitution effect of earned income is negative, thus decreasing the retirement probability. The retirement decision is strongly state persistent for up to three years after the initial retirement decision and the state persistence of retirement is reinforced by wealth and earned income.