majestickhan / nowcasting-python Goto Github PK

Python Nowcasting

License: BSD 3-Clause "New" or "Revised" License

Python 100.00%

nowcasting economics macroeconomics python-nowcasting forecast federal-reserve-bank dfm

nowcasting-python's Introduction

Nowcasting-Python

The repository contains Python code that is translated from a Matlab code which produces a dynamic factor model. The Matlab code and the model belong to the Federal Reserve Bank of New York, developed by Eric Qian and Brandyn Bok. Please visit their repository for further details.

The Matlab code being translated implements the nowcasting framework described in "Macroeconomic Nowcasting and Forecasting with Big Data" by Brandyn Bok, Daniele Caratelli, Domenico Giannone, Argia M. Sbordone, and Andrea Tambalotti, Staff Reports 830, Federal Reserve Bank of New York (prepared for Volume 10 of the Annual Review of Economics).

File and folder description

data/ : example US data downloaded from FRED
Functions/ : functions for loading data, estimating model, and updating predictions
example_DFM.py : example script to estimate a dynamic factor model (DFM) for a panel of monthly and quarterly series
example_Nowcast.py : example script to produce a nowcast or forecast for a target variable, e.g., real GDP growth
ResDFM.pickle : example DFM estimation output
Spec_US_example.xls : example model specification for the US

NOTICE:

This repository is not associated Federal Reserve Bank of New York. I am not well versed in dynamic factor modeling but the reason I created this repository was to challenge myself and I thought it would be cool to convert a model from matlab to python. So feel free to use it for academic purposes but make sure you give proper credit to Eric Qian and Brandyn Bok.

The repository is in its early stages and requires further improvements such as optimizing functions and writing clearer syntax with proper formatting. I'm not sure when I will get to this as I moved away from time series and started focusing on deep learning. My apologies for any inconviences. Feel free to post any technical issues that you encountered and I'll try my best to respond promptly.

Thanks,

nowcasting-python's People

Contributors

Stargazers

Watchers

nowcasting-python's Issues

Next steps

Unit testing

Create unit test to ensure python model is working properly

Graphs and formatting

Plotting graphs: finished plotting graphs for example_dfm.py in plot
Plot loglik in Plotly
Branch plot has been merged to the master. Should be able to see the graphs that are show in the Matlab code.

Comments:

Next set of tasks is working on creating unit tests

About update_Nowcast.py

Thanks for sharing your code.
Recently I'm learning about nowcasting models, your code is just what I need.
But I got something wrong when running your update_Nowcast.py，so I tried to fix it.

When I change the vintage, I may got error on line 407 ~ line 422，so I rewrite like this

for jk in range(1, lag + 1):

    Plag.append(np.zeros(Plag[0].shape))

    for jt in range(Plag[0].shape[0], lag, -1):

        As = np.matmul(np.matmul(Vf[jt-jk-1], A.T), np.linalg.pinv(np.matmul(np.matmul(A, Vf[jt-jk-1]), A.T) + Q))

        Plag[jk][jt-1] = np.matmul(As, Plag[jk-1][jt-1])

Also you mention that weight seems suspicious.
I find a mistake on line 281，it should be
if t_miss[j] > t_miss[i]:

I'm not sure whether line 245 should like this
m = max(t_miss[i],t_fcst[0])

Finally, I run the code with vintage_old = '2016-11-30' and vintage_new = '2016-12-23'.
I can get these：

          2016-11-30 nowcast:              2.80471
  Impact from data revisions:      0.10666
   Impact from data releases:      -0.49701
                                 +_________
                Total impact:      -0.39035
          2016-12-23 nowcast:              2.41436

Is there any way that I can use your code with yealy data?

thanks for your code! I find freqs with [day, week, month,quarter, year] in your code, but the example code just use month and quarter data. Is there any way that I can use your code with yearly data?

Release Dates

Hey,

I dont know if this is an issue with how I am formatting data or something else, but I am trying to run the code on dataframes which have different release dates.

As a result I have formatted the data in a weekly format with NAs in between.

When I run this code i then either get the error:

ValueError: expected square matrix

ValueError: Internal work array size computation failed: -5

Ideally I would also be able to run with actual weekly data as an input.

Have attached the relevant files and would love any thoughts you have?
Spec_US_example.xlsx

2016-06-29.xlsx

majestickhan / nowcasting-python Goto Github PK

nowcasting-python's Introduction

Nowcasting-Python

File and folder description

NOTICE:

nowcasting-python's People

Contributors

Stargazers

Watchers

Forkers

nowcasting-python's Issues

Next steps

Unit testing

Graphs and formatting

Comments:

About update_Nowcast.py

Is there any way that I can use your code with yealy data?

Release Dates

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent