李子奈教科书第二版p86-p87的学习笔记：

/* 1、 Chow 模型稳定性检验（lrtest） */
* 用似然比作chow检验，chow检验的零假设：无结构变化，小概率发生结果变化
* 估计前阶段模型
qui reg lnq lnx lnp0 lnp1 in 1/14
est store A
* 估计后阶段模型（1995-2001）
qui reg lnq lnx lnp0 lnp1 in 15/21
est store C
* 整个区间上的估计结果保存为All
qui reg lnq lnx lnp0 lnp1 in 1/21
est store All
* 用似然比检验检验结构没有发生变化的约束
lrtest (All)(A C),stats

/* 2、Chow 模型稳定性检验（F-test） */
* 用F-test作chow间断点检验检验模型稳定性，p86
* chow检验的零假设：无结构变化，小概率发生结果变化
* 估计前阶段模型
qui reg lnq lnx lnp0 lnp1 in 1/14
scalar n1=e(N)
scalar rss1=e(rss)
* 估计后阶段模型（1995-2001）
qui reg lnq lnx lnp0 lnp1 in 15/21
scalar n2=e(N)
scalar rss2=e(rss)
* 整个区间上的估计结果保存为All
qui reg lnq lnx lnp0 lnp1 in 1/21
scalar k=e(df_m)
scalar rssr=e(rss)
* 用F检验检验结构没有发生变化的约束
*计算和显示 F检验统计量p.85顶部公式，零假设：无结构变化
scalar f_test=(rssr-(rss1+rss2))/(k+1)/(rss1+rss2)*(n1+n2-2*(k+1))
dis f_test
* F统计量的临界概率
dis Ftail((k+1),(n1+n2-2*(k+1)),f_test)
* F统计量的临界值
dis invFtail((k+1),(n1+n2-2*(k+1)),0.05)

/* 3、F检验实现Chow预测检验检验模型的稳定性 */
*计算和显示 F检验统计量 ，p87，零假设：无结构变化
scalar ff_test=(rssr-rss1)/n2/(rss1)*(n1-k-1)
dis ff_test
* F统计量的临界概率
dis Ftail(n2,(n1-k-1),ff_test)
* F统计量的临界值
dis invFtail(n2,(n1-k-1),0.05)

http://www.stata.com/support/faqs/stat/chow3.html

http://www.stata.com/support/faqs/stat/chow.html

http://www.stata.com/support/faqs/stat/awreg.html

Test for a shift in regression coefficients                      (STB-17: sts7)
-------------------------------------------

        chow varlist [weight] [if exp] [in range] [, chow(sample-list)
                        noconstant current(varlist) detail exclude(varlist)
                        lags(#[,#[,...]]) preserve regress restrict(varlist)
                        nosample static(varlist) time(power) vartest

chow calculates the Chow and Farley-Hinich-McGuire tests for a shift
in regression coefficients

Options
-------

chow(sample-list) specifies the subsamples for the Chow test.

noconstant suppresses the constant.

current(varlist) names the time series variables whose contemporaneous
    values are to be included as regressors.

detail indicates that each subsample is to be tested separately.

exclude(varlist) excludes variables from the Chow test.

lags(#[,#[,...]) specifies the lags.

preserve preserves the original data set.

regress displays the Chow regression.

restrict(varlist) holds the coefficients of the variables in the
    varlist constant across the subsamples.

nosample suppresses the display of sample coverage information.

static(varlist) specifies the static variables.

time(power) indicates the function of time to use for a Farley-
    Hinich-McGuire test.  Instead of a number, you also may specify
    "exp", "ln", or "log".

vartest requests a test of the hypothesis that the error variance is
    equal across subsamples.

Examples
--------

        chow y x1 x2, chow(year>1975)

tests whether the coefficients of the regression of y on x1 and x2 change
after 1975.

        chow gnp money, chow(year>1982) lags(4) restrict(gnp)

tests whether the coefficients on four lags of money change after 1982.

Author
------
        Sean Becketti
        Stata Technical Bulletin

Also see
--------

    STB:  sts7 (STB-17)
On-line:  help for ts