공적연금제도의 실시와 더불어 가장 큰 논란의 대상 하나는 공적연금제도의 도입이 민간부문의 소비행태, 즉 저축에 어떤 영향을 미치느냐는 것이다. 본 연구에서는 부과방식인 미국의 공적연금제도(OASI: Old Ages Survivors Insurance)가 미국의 민간부문 저축에 미치는 효과를 준모수적 방법(Semi-parametric Method)을 이용하여 분석하였다. 실증분석에 대한 기존의 연구와 달리 준모수적 방법의 비모수적 부분은 모형 설정에 대한 사전적인 지식을 필요로 하지 않는다. 사회보장부가 소비에 미칠 효과를 데이터에만 의존하는 비모수적 방법을 사용하는 경우 1992년 현재 OASI가 미국의 민간저축을 6～13.45% 감소시켰다는 추정결과를 얻었다. 이는 민간저축의 59%를 감소시켰다는 펠트스타인(1996)의 실증분석 결과가 비모수적 방법에 의해서는 강하게 뒷받침되지 않음을 시사한다.

During the past decades, debate has been focused on whether or not the United States Social Security program(OASI) has depressed private saving in the economy. In the context of concern about low levels of saving, capital formation and the consequent impact on productivity growth and output, the issue is clearly important. In a pioneer article, Martin Feldstein(1974) estimated that the introduction of the Social Security System had reduced personal saving in the United States by 50 percent. Subsequent empirical studies presented estimates both supporting and contradicting Feldstein`s conclusion. The debate is still inconclusive. Recently, Feldstein(1996) reexamined the results of his 1974 paper on Social Security and saving. His new estimates imply that the Social Security program currently reduces overall private saving by nearly 60 percent in the aggregate. While his estimation is based on the traditional life cycle model that includes the SSW(Social Security Wealth) variable as the constructed variable, the SSW variable is included as the linear form in the model. However, the estimated value of the SSW might be sensitive on the functional form when the SSW variable enters the consumer expenditure form. Remembering that Feldstein`s estimation results could be sensitive to functional form, the non-parametric methodology, which is model specification free, was used to estimate the SSW variable. Semi-parametric estimation shows that the current U.S. Social Security program has reduced overall private saving by between 6.9 and 13.45 percent in the aggregate. This result doesn`t strongly support Feldstein`s(1996). It is noted, however, that this result is not robust in that the estimated values are sensitive to the optimal bandwidth.