259 data points with goods=169 and bads=90 should be able do your analysis. Now you have all data points of 全部中小企业, it IS population data. There is no sampling here. So the bootstrap is useless. There is no needs to have 自然配比=1.There are many logistic analyses in which the research interesting is only the treatment effect rather that the probablity of being bads. When 自然配比=1, the biased estimator is the constant term and hance the probablity of being bads/goods(only true under logistics assumptions). It can be easy proved by a small simulation programs. The therotical proof can be found in
page 90 of Limited-dependent and qualitative variables in econometrics by Maddala. Actually the proof is easy only involving the math of a middle-school.

The advantage 自然配比=1 is
1) reduce cost   --- expecially in medical research
2) the efficient loss is much less than a random sample  --- events contains more information than nonevents becuase the number of events is much less than that of nonevents.
3) it is less efficient than using all data. --- this is simple true because sample with  自然配比=1 is a sub-sample of all data.

These points are useless in your case.

I would suggest that,
1) build a model with all data
2) build a model with data up to 2008

Contrasting these two models it may give you some hints about your models.   

The data quality may need to pay more attension rather than quantity which you can do nothing about it.

Hope this will shed some lights on your research project.

不是二阶的方法，是二分变量的方法，其实不用考虑太复杂了。你参考一下古扎拉蒂的中级第15章的定性响应回归模型就可以了。只不过这是非群组的数据，采用个体的Logistic回归就可以。
另外推荐参考书：王济川 郭志刚 《Logistic 回归模型——方法与应用》，高等教育出版社，十分详细的讲述了估计、解释、诊断和评价的内容。
Eviews软件中有采用make equation 中method 选择Binary，再选择的Logistic选项就可以。
祝好运！

2# 三攀

呵呵，谢谢啦。但是，我的问题其实不是Logistic回归二值变量的问题，而是怎么能充分利用其余100多个的样本点。有点类似Bootstrap那种方法，但是又不是靠仿真模拟，而是一种计量方法。呵呵，我明天先去图书馆查查你推荐的书

2# 三攀
三攀，你好
    学校图书馆放假，下周一才能开。不过，我在Greene的《计量经济分析》里看到了响应回归相关内容。里面介绍了bootstrap和选择抽样法（choice based sampling——加权内生抽样极大似然估计量：weighted endogenous sampling maximun likelihood）。前者解决的是样本不足的问题，后者解决的是样本中0和1比例与总体中0、1配比不同的问题,采用给样本中“比例被低估的部分 ”加权的方法，但是似乎只能手算，没有成熟的软件。
   你说的用Eviews的方法，我觉得那是不是只是做普通的Logistics回归，还是没有解决样本配比与实际总体配比不同的问题？
   呵呵，麻烦啦

There is several things you need to sort out before you analyze your data.

1) Is that a random sample data. If it is NOT a random sample, then you really need to understand  sampling process. This will bised all you analysis.For example, the data is sample in beijing, but you want to infer all 中小企业违约 in China. Or the data you got is from ONE particular bank, it may ONLY represent that bank clients' behevior nothing else.

2) If your purpose is to estimate the probability of 中小企业违约, then the sampling weight needs to be know. Because you may oversampling the case of 1(bads/events) and undersampling the case of 0(goods/nonevebts). In this case the probability of 中小企业违约 is for the in-sample data NOT for the population. In theory, the logistic regression bised the intercept only when you only stratify the response in your sampling process.

3) In your case particular, you should use all your data. There is no need to farther sample down.

Sample down happens only,
1) cost of cellect data  -- it is irrelavent in this case.
2) easy for modelling  -- if you have many data say 1 million cases. This is not the case here.

Hope this helps.