slide sfdp rotterdam_2014_june

13

Click here to load reader

Upload: xi-hao-li

Post on 10-Jul-2015

107 views

Category:

Presentations & Public Speaking


0 download

DESCRIPTION

Presentation in 34th International Symposium on Forecasting, Rotterdam, June 29 - July 2, 2014

TRANSCRIPT

Page 1: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection

Stock-Flow Dynamic Projection

Mauro Gallegati Xihao Li

Department of Economics and Social Sciences (DiSES)

Universita Politecnica delle Marche

June 30, 2014

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 2: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Introduction: I

Consider the Era of Big Data:

Economic Entities (firms/banks) provide accountingstatements for reporting: the balance sheet, theincome statement, the statement of cash flows;

in a faster pace, not only annual reporting but alsopossibly quarterly or monthly reporting.

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 3: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Introduction: II

Question: Can we take advantage of this new stream ofeconomic data?

to improve our capability of forecasting andmonitoring macroeconomic fluctuation, or evencrisis?

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 4: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Idea: I

Experience from agent-based economic research:economic agents’ micro-level interaction leads tostructural transition in meso-level which results inmacro-level economic fluctuation.

Two types of economic variables:

1 stock variable that measures quantities at a timepoint, e.g. firms’ equity in the balance sheet;

2 flow variable that measures quantities at a timeinterval, e.g. firms’ revenue in the income statement.

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 5: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Idea: II

Main idea of Dynamic projection:

Macro-level economic variable = aggregation inmicro-level( stock / flow variable ) + aggregation inmeso-level( the impact of interaction amongeconomic entities ) (∗)

Assume ’as-if’ the economy in the future ceterisparibus, compute dynamic projection for the futurestate of macro-level economic variable by usingmicro-level stock and flow data to measure eachcomponent in formula (∗) .

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 6: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Example

Use the dataset of Japanese firm’s financial statements 1:

for 4599 firms listed in Tokyo Stock Exchange

for 33 years of annual financial statements, i.e.balance sheet, profit and loss statement(PLstatement)2, from the year of 1980 to 2012.

1To use this dataset, the authors acknowledge the support from the European Community Seventh

Framework Programme (FP7/2007-2013) under Socio-economic Sciences and Humanities, grantagreement no. 255987 (FOC-II)

2Profit and loss statement is equivalent to income statement.

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 7: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Example

Consider the following target variables:

stock variable: aggregate equity A from balancesheet

flow variable: aggregate gross profit π from PLstatement.

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 8: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Example

Dynamic projection for one-period-ahead out-of-sampleforecast, to compare with the benchmark of ARIMA:

1 Use data for period of 1980 to 1996 as initialinformation set,

2 at the end of each period t = 1996, ..., 2011, computeone-period-ahead out-of-sample forecast for X = A,or π:

dynamic projection:dp(X )t+1|t = {dp(X )1997|1996, . . . , dp(X )2012|2011}choose the optimal ARIMA according to theBIC(AICc) information criterion, then use the optimalARIMA to conduct the forecast:ARIMA(X )t+1|t = {ARIMA(X )1997|1996, . . . , ARIMA(X )2012|2011}

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 9: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Example

One-period-ahead out-of-sample forecast: aggregate equity

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

120

140

160

180

200

220

240

260

year

aggr

egat

e eq

uity

forecast: ARIMA Vs. dynamic projection

realizationARIMAdynamic projection

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012−20

−10

0

10

20

year

aggr

egat

e eq

uity

forecast error: ARIMA Vs. dynamic projection

ARIMAdynamic projection

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 10: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Example

One-period-ahead out-of-sample forecast: aggregate profit

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

40

45

50

55

60

65

70

75

80

85

90

year

aggr

egat

e gr

oss

prof

it

forecast: ARIMA Vs. dynamic projection

realizationARIMAdynamic projection

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

−10

−5

0

5

10

year

aggr

egat

e gr

oss

prof

it forecast error: ARIMA Vs. dynamic projection

ARIMAdynamic projection

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 11: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Example

Diebold-Mariano test.

Null hypothesis: comparing with dp(X )t+1|t ,

ARIMA(X )t+1|t has the same or higher accuracy inforecasting, for X = A, or π.

use linear loss function and quadratic loss function.

for aggregate equity A:p-value Vs. ARIMA with BIC Vs. ARIMA with AICc

Power = 1 0.036 0.028Power = 2 0.042 0.029

for aggregate gross profit π:p-value Vs. ARIMA with BIC Vs. ARIMA with AICc

Power = 1 0.014 0.022Power = 2 0.027 0.033

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 12: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Concluding Remark

In our story of aggregate equity and aggregateprofit, dynamic projection shows higher accuracy inone-period-ahead out-of-sample forecast thanARIMA.

Is pure luck or any theory behind?

Working in progress: mathematical inference frommulti-level dynamical system. 3

3See the MatheMACS project, supported under ”ICT-2011.9.7 FET Proactive: Dynamics of Multi-Level

Complex Systems (DyM-CS)”, http://www.mathemacs.eu/.

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam

Page 13: Slide sfdp rotterdam_2014_june

Stock-Flow Dynamic Projection Introduction Idea Example Concluding Remark

Thank you!

Mauro Gallegati, Xihao Li 34th International Symposium on Forecasting, Rotterdam