DSGE动态经济学建模Band-Pass filter 程序分享-经管之家官网！

这个代码我调试过，添加了缺少的函数文件 ols.m ，但是依然还有10% 的错误，本来用markdown 发帖，但是贴了主程序1100行的代码，太大导致无法发出去。 所以就算了。
这里插入一个函数计算 band-pass filter的。

1. function yf=BPF(y,up,dn,K)
   
2. 
   
3. % bpf.m
   
4. % Program to compute band-pass filtered series
   
5. % Inputs are
   
6. %y: data (rows = observations, columns=series)
   
7. %up:period corresponding to highest frequency (e.g., 6)
   
8. %dn:period corresponding to lowest frequency (e.g., 32)
   
9. %K: number of terms in approximating moving average
   
10. %[calls filtk.m (filter with symmetric weights) as subroutine]
    
11. x=[up dn];
    
12. disp(' ')
    
13. disp('bpf(y,up,dn,K): band-pass filtering of series y with symmetric MA(2K+1)')
    
14. disp(' ')
    
15. disp(' for additional information see: ')
    
16. disp(' ')
    
17. disp(' M. Baxter and R.G. King ')
    
18. disp(' ')
    
19. disp(' Measuring Business Cycles: ')
    
20. disp(' Approximate Band-Pass Filters')
    
21. disp(' for Macroeconomic Time Series')
    
22. disp(' ')
    
23. disp('Filter extracts components between periods of: ')
    
24. disp(' up dn')
    
25. disp(x)
    
26. % pause(2)
    
27. 
    
28. if (up>dn)
    
29. disp('Periods reversed: switching indices up & dn')
    
30. disp(' ')
    
31. dn=x(1); up=x(2);
    
32. end
    
33. 
    
34. if (up<2)
    
35. up=2;
    
36. disp('Higher periodicity > max: Setting up=2')
    
37. disp(' ')
    
38. end
    
39. 
    
40. % convert to column vector
    
41. [r c]=size(y);
    
42. if (r<c)
    
43. y=y';
    
44. disp('There are more columns than rows: Transposing data matrix')
    
45. disp(' ')
    
46. end
    
47. 
    
48. % Implied Frequencies
    
49. omubar=2*pi/up;
    
50. omlbar=2*pi/dn;
    
51. 
    
52. % An approximate low pass filter, with a cutoff frequency of "ombar",
    
53. % has a frequency response function
    
54. %
    
55. %alpha(om) = a0 + 2*a1 cos(om) + ... 2*aK cos(K om)
    
56. %
    
57. % and the ak's are given by:
    
58. %
    
59. % a0 = ombar/(pi) ak = sin(k ombar)/(k pi)
    
60. %
    
61. % where ombar is the cutoff frequency.
    
62. 
    
63. % A band-pass filter is the difference between two
    
64. % low-pass filters,
    
65. % bp(L) = bu(L) - bl(L)
    
66. % with bu(L) being the filter with the high cutoff point and bl(L) being
    
67. % that with the low cutoff point.Thus, the weights are differences
    
68. % of weights for two low-pass filters.
    
69. 
    
70. % Construct filter weights for bandpass filter (a(0)....a(K)).
    
71. 
    
72. akvec=zeros(1,K+1);
    
73. 
    
74. akvec(1)=(omubar-omlbar)/(pi);% weight at k=0
    
75. 
    
76. for k=1:K;
    
77. akvec(k+1)=(sin(k*omubar)-sin(k*omlbar))/(k*pi); % weights at k=1,2,...K
    
78. end
    
79. 
    
80. % Impose constraint on frequency response at om = 0
    
81. % (If high pass filter, this amounts to requiring that weights sum to zero).
    
82. % (If low pass filter, this amounts to requiring that weights sum to one).
    
83. 
    
84. if (dn>1000)
    
85. disp('dn > 1000: assuming low pass filter')
    
86. phi=1;
    
87. else
    
88. phi=0;
    
89. end
    
90. 
    
91. % sum of weights without constraint
    
92. theta=akvec(1)+2*sum(akvec(2:K+1));
    
93. % amount to add to each nonzero lag/lead to get sum = phi
    
94. theta=(phi-theta)/(2*K+1);
    
95. % adjustment of weights
    
96. akvec=akvec+theta;
    
97. 
    
98. % filter the time series
    
99. yf=filtk(y,akvec);
    
100. 
     
101. if (r<c)
     
102. yf=yf';
     
103. end
     
104. 

复制代码
插入附件，特别大也许会消耗论坛流量，因为有143M 这样大。 文件夹里有数据、主程序脚本文件、函数文件。 希望对大家的学习有帮助，未来三年推动国内的动态宏观经济学建模向国际一流接近。
因为太大所以我用了网盘，发链接吧。
链接：https://pan.baidu.com/s/1XyKiuoi5g5r_8L2Qregjhw
提取码：取消密码公布