## 6. Forecasting Manufacturing Value Added in 2009 Using IPI: a Comparison Between PFD and EPFD

## Source details

**Canonical URL:** [6. Forecasting Manufacturing Value Added in 2009 Using IPI: a Comparison Between PFD and EPFD](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12169.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12169.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12169.pdf.json)

---

### I. Introduction: benchmarking and extrapolation
- Benchmarking problem split into distribution and extrapolation.
- Basic Denton PFD (Denton, 1971) with Cholette (1984) starting-condition modification aims to make proportionate period-to-period changes of the benchmarked series as close as possible to those of a preliminary series.
- QNA Manual (Bloem et al., 2001) proposes an enhanced Denton PFD for extrapolation that incorporates a user forecast of the annual benchmark-to-indicator (BI) ratio for the next year to align quarterly extrapolations with that annual BI forecast.
- This paper:
  - formalizes the enhanced Denton PFD in matrix form,
  - derives an analytical solution,
  - compares it to both the QNA Manual shortcut and the basic Denton formula using artificial and real-life examples.

### II. The basic Denton PFD benchmarking method (formulation and properties)
- Problem setup:
  - y: (1)sN× vector of unknown quarterly benchmarked values.
  - y0: (1)N× vector of known aggregated (annual) values.
  - p: (1)sN× vector of preliminary values with Jp ≠ 0.
  - Temporal aggregation constraint: 0 = J y, with J = I_s ⊗ 1′ and s (e.g., s = 4 for quarterly-to-annual).
- Objective (Cholette modified Denton PFD): minimize sum of squared proportional first differences between y and p subject to aggregation constraint. Matrix form: minimize (ŷ − p)′ Q (ŷ − p) s.t. J y = y0 with Q = (D′ diag(p)^{-1} D)^{-1} and D the first-differences matrix.
- Extrapolation property of basic Denton PFD:
  - For extrapolated periods, BI ratio corresponds to the BI ratio of the fourth quarter covered by the last benchmark year.
  - Quarterly growth rates of extrapolated y will match quarterly growth rates of p, implying the annual growth rate of y for year N+1 is implicitly derived from the last available BI ratio and user has no control.

### III. The enhanced Denton PFD method for extrapolation (analytical solution)
- Added constraint: impose a user forecast of the annual BI ratio for year N+1:
  - Constraint (5): weighted sum of quarterly benchmarked values for year N+1 equals forecast BI ratio times aggregated preliminary series for year N+1.
  - BI forecast parameterization: b_{N+1} = q b_N where q is expected forecast change (expression (8)).
- Matrix implementation:
  - Added constraint expressed in matrix form (expression (9)); combined constraint system in (10).
  - Distributed and extrapolated values obtained by using formula (4) with modified J* and y0*.
- Advantages of the analytical enhanced solution:
  - Explicitly incorporates user forecast of annual BI ratio for next year.
  - Offers control over annual extrapolated BI ratio and reduces potential bias in purely mechanical Denton extrapolation.
  - Analytical solution differs quarter-by-quarter from both basic Denton and the QNA Manual shortcut; solves the enhanced constrained optimization exactly.

### III.A. QNA Manual shortcut approximation of the enhanced method
- Shortcut constructs adjustment constant η defined by η = (q b_{N+1} − q_{sN−1}) / 3 (expression (11)) to adjust quarterly BI ratios beginning from second quarter of last available year.
- Recursion to compute extrapolated quarterly BI ratios using η provided.
- Properties:
  - Shortcut guarantees the implicit annual extrapolated BI ratio equals the user-requested annual BI ratio.
  - Shortcut provides suboptimal quarter-to-quarter movements relative to the analytical solution.

### IV. Artificial-data example (QNA Manual example 6.2 reproduced)
- Setup specifics used in the example:
  - Last available quarterly BI ratio 1999:Q4 = 10.355.
  - User forecast: BI ratio in 2000 increases 2.0 percent from previous year: resulting BI ratio 10.486 = 10.280 × 1.02.
  - Shortcut η = (10.355 − 10.486) / 3 = −0.044.
- Key numeric outcomes:
  - Indicator annual growth for 2000: 0.9 percent.
  - Basic Denton PFD annual rate for 2000: 1.6 percent.
  - Enhanced Denton PFD (shortcut) annual rate for 2000: 2.9 percent (difference from indicator growth exactly equal to 2.0 percent forecast change in BI ratio).
  - Analytical enhanced solution yields the same annual rate for 2000 as the shortcut (2.9 percent) but different quarter-to-quarter values and growth rates.
- Mean squared differences (MSD) of quarterly growth rates (square root of average of squared differences between growth rates of indicator series and benchmarked series):
  - MSD total (entire sample): Basic Denton PFD 0.5946, Enhanced PFD shortcut 0.6523, Enhanced PFD analytical 0.6392.
  - MSD year 2000 (extrapolated quarters 2000:Q1–2000:Q4): Basic Denton PFD 0.0000, Enhanced PFD shortcut 0.4184, Enhanced PFD analytical 0.3312.
- Interpretation:
  - Basic Denton preserves indicator quarterly movements perfectly in extrapolation (hence MSD 0.0000 for extrapolated year).
  - Analytical enhanced is closer to indicator movements than the shortcut in extrapolation, but both enhanced methods achieve the requested annual BI ratio for 2000 (2.9 percent).
- Sensitivity over b_{2000} ∈ [0.94, 1.06]:
  - Analytical enhanced method always closer to indicator movements (lower PFD criterion and MSD) than shortcut.
  - Improvement of analytical over shortcut larger as BI ratio change deviates more from 1.0.

### V. Application to real-life data: forecasting manufacturing VA for 2009 using IPI
- Objective: out-of-sample extrapolation of manufacturing value added (VA) for 2009 using quarterly industrial production index (IPI) of manufacturing as the indicator series via Denton PFD methods.
- Rationale: IPI is a timely indicator often used as a short-term proxy for VA; 2008–2009 saw a strong downturn breaking many historical short-term relationships.
- Data and sample:
  - Countries: Australia, Canada, France, Germany, Indonesia, Italy, Japan, Korea, Mexico, South Africa, Turkey and the United States (twelve G-20 countries).
  - Sources: time series extracted from “National Accounts” and “Industry and Services” themes in the OECD data warehouse (http://stats.oecd.org/).
  - Sample period: 2000 through 2009.
  - Extraction date: March 2012.
- Expectation: enhanced Denton PFD may improve extrapolation for 2009 relative to basic Denton PFD if the break in the IPI–VA relationship is properly addressed by incorporating a forecasted annual BI ratio.

### VI. Methodology and problem statement (Section II reiteration)
- With the basic Denton PFD method, the last available quarterly BI ratio is carried forward to extrapolate subsequent quarters not covered by annual benchmarks:
  - Quarterly benchmarked estimates will show the same quarterly growth rates as the indicator series.
  - The annual growth rate may differ from that shown by the indicator series.
- Enhanced Denton requires an input: forecast of the annual BI ratio for the next year (see equation (7) in source).
- Out-of-sample exercise: assume BI ratio for 2009 unknown; derive forecast for 2009 given BI ratios up to and including 2008.

### VII. BI ratio behavior and forecasting rules
- Observed patterns (2000–2008) justify simple forecasting rules given low number of observations.
- Interpretation of BI ratio movements:
  - Increase of BI ratio over time: VA growing faster than IPI.
  - Decrease: VA growing slower than IPI.
  - Stable around a mean: IPI is a good approximation of VA growth.
- Forecasting rules used to derive annual 2009 BI forecast (using 2000–2008 data):
  - If BI ratio fluctuates around a long-term average, use the average level of the BI ratio observed in 2004–2008. Used for: Australia, Italy, Mexico.
  - If BI ratio displays a clear trend in latest years, use the geometric average of the growth rates of the BI ratio observed in 2004–2008. Used for: Canada, France, Germany, Indonesia, Japan, Korea, South Africa, Turkey, United States.
- Rationale: five-year span is a subjective choice aiming for a simple, pragmatic forecasting method to improve extrapolation quality relative to basic Denton.

### VIII. Extrapolation experiment and comparative results (Table 6 summary)
- Observations:
  - In most countries IPI 2009 % growth was more negative than VA (Y GDP) 2009 % growth (exceptions: Germany, Korea, Mexico).
  - Basic Denton PFD tends to produce 2009 extrapolated VA growth rates similar to the IPI growth rate but does not incorporate historical BI ratio movements.
  - Enhanced Denton PFD incorporates the user-provided annual BI forecast and uses historical BI movements to improve extrapolation.
- Country-level results (columns interpreted as: IPI 2009 % growth ; Y GDP 2009 % growth ; Basic Denton PFD Q GDP 2009 % growth ; Basic Denton 2009 error ; Enhanced Denton PFD Q GDP 2009 % growth ; Enhanced Denton 2009 error):
  - Australia*: IPI -7.3 ; Y GDP 1.5 ; Basic Denton Q GDP -9.4 ; Basic error -10.9 ; Enhanced Q GDP -2.3 ; Enhanced error -3.8
  - Canada: IPI -13.9 ; Y GDP -13.6 ; Basic Denton Q GDP -13.9 ; Basic error -0.3 ; Enhanced Q GDP -13.8 ; Enhanced error -0.2
  - France: IPI -13.9 ; Y GDP -11.0 ; Basic Denton Q GDP -13.7 ; Basic error -2.7 ; Enhanced Q GDP -13.5 ; Enhanced error -2.5
  - Germany: IPI -17.0 ; Y GDP -18.1 ; Basic Denton Q GDP -17.8 ; Basic error 0.3 ; Enhanced Q GDP -18.2 ; Enhanced error -0.1
  - Indonesia: IPI 1.4 ; Y GDP 2.2 ; Basic Denton Q GDP 1.8 ; Basic error -0.4 ; Enhanced Q GDP 3.8 ; Enhanced error 1.6
  - Italy*: IPI -19.3 ; Y GDP -16.0 ; Basic Denton Q GDP -19.5 ; Basic error -3.5 ; Enhanced Q GDP -18.7 ; Enhanced error -2.7
  - Japan: IPI -21.9 ; Y GDP -19.7 ; Basic Denton Q GDP -21.4 ; Basic error -1.7 ; Enhanced Q GDP -20.6 ; Enhanced error -0.9
  - Korea: IPI -0.9 ; Y GDP -1.5 ; Basic Denton Q GDP -1.0 ; Basic error 0.5 ; Enhanced Q GDP -1.3 ; Enhanced error 0.2
  - Mexico*: IPI -9.5 ; Y GDP -9.9 ; Basic Denton Q GDP -9.5 ; Basic error 0.4 ; Enhanced Q GDP -9.9 ; Enhanced error 0.0
  - South Africa*: IPI -12.6 ; Y GDP -10.4 ; Basic Denton Q GDP -12.2 ; Basic error -1.8 ; Enhanced Q GDP -11.2 ; Enhanced error -0.8
  - Turkey: IPI -11.5 ; Y GDP -7.2 ; Basic Denton Q GDP -11.1 ; Basic error -3.9 ; Enhanced Q GDP -10.1 ; Enhanced error -2.9
  - United States: IPI -13.5 ; Y GDP -9.7 ; Basic Denton Q GDP -13.8 ; Basic error -4.2 ; Enhanced Q GDP -12.5 ; Enhanced error -2.8
- Aggregate error statistics:
  - Mean error (Basic Denton PFD): -2.3 percent
  - Mean error (Enhanced Denton PFD): -1.2 percent
  - Mean absolute error (Basic Denton PFD): 2.5
  - Mean absolute error (Enhanced Denton PFD): 1.6
- Notes:
  - *For Australia, Italy, Mexico, the annual BI forecast required by the enhanced method is the average level of the BI ratios in 2004–2008.
  - For other countries, the annual BI forecast is the geometric average of the annual growth rates of the BI ratios in 2004–2008.

### IX. Findings, interpretation and methodological implications
- Basic Denton PFD:
  - Preserves quarterly movements of the indicator in extrapolation (no quarter-to-quarter adjustment), which yields zero MSD for extrapolated year but offers no user control of annual BI ratio.
  - Extrapolation is purely mechanical and may miss systematic bias between indicator and target.
- Enhanced Denton PFD:
  - Allows explicit imposition of a forecast for annual BI ratio b_{N+1} (e.g., via multiplicative factor q relative to b_N).
  - Analytical solution (matrix formulation) provides exact constrained optimization incorporating the BI forecast.
  - Analytical enhanced solution outperforms the QNA Manual shortcut in terms of closeness to indicator movements (lower PFD criterion and MSD), particularly when the requested BI ratio change is larger (further from 1.0).
  - Shortcut guarantees the requested annual BI ratio but yields suboptimal quarter-to-quarter preservation relative to analytical enhanced solution.
- Practical implications for national accounts:
  - Use of enhanced Denton PFD with an analytically derived solution gives users control over anticipated annual BI ratio while preserving quarterly continuity and better matching indicator movements than the shortcut.
  - When the annual benchmark for year N+1 becomes available and matches the BI ratio forecast used in extrapolation, enhanced Denton PFD will reduce future revisions of quarterly data.
- Caveat:
  - Effectiveness of enhanced method depends on quality of annual BI forecast; poor forecasts (e.g., Indonesia in this exercise) can make enhanced method perform worse than basic method.

### X. Conclusions and recommendations
- Limitations of basic Denton PFD:
  - Implicit annual forecast for the next year is effectively determined by the level of the last observed annual BI ratio and may be biased by preliminary information.
- Recommendation:
  - Prefer the analytical matrix solution of the enhanced Denton PFD method presented in this paper rather than the QNA Manual’s shortcut recursion, because the shortcut may distort quarter-to-quarter movements in the indicator series.
  - When a systematic pattern arises from the annual BI ratio series—especially in the latest years—users should exploit this information by applying the enhanced Denton PFD method.
- Final caveat:
  - Performance gains from the enhanced method require reasonable BI ratio forecasts; poor forecasts can reverse the expected gains (as illustrated for Indonesia).

*Source: IMF Working Paper — section "6. Forecasting Manufacturing Value Added in 2009 Using IPI: a Comparison Between PFD and EPFD" (text and tables as provided).*

### References .............................................................................................................

### _wp12169 - References .............................................................................................................

### Tables
- 1.  Extrapolation Using Forecast BI Ratios (Example 6.2, QNA Manual, 2001) ....................10
- 2.  Enhanced Denton PFD: Comparison Between the Shortcut and the Analytical Solution ..12
- 3.  Enhanced Denton PFD: Comparison with the Indicator Series ..........................................12
- 4.  Basic Denton PFD vs. Enhanced Denton PFD: MSD of Quarterly Growth Rates .............12
- 5.  Enhanced Denton PFD: Comparison Between the Analytical Solution and the Shortcut 
  Version with Different BI Ratios .............................................................................................13

*Source: _wp12169 - References .............................................................................................................*

### 6.  Forecasting Manufacturing Value Added in 2009 Using IPI: a Comparison Between

### 6.  Forecasting Manufacturing Value Added in 2009 Using IPI: a Comparison Between PFD and EPFD

### I. Introduction: benchmarking and extrapolation
- Benchmarking problem split into distribution and extrapolation.
- Denton (1971) PFD (Proportional First Differences) method with Cholette (1984) starting-condition modification (the basic Denton PFD) aims to make proportionate period-to-period changes of the benchmarked series as close as possible to those of a preliminary series.
- QNA Manual (Bloem et al., 2001) proposes an enhanced Denton PFD for extrapolation that incorporates a user forecast of the annual benchmark-to-indicator (BI) ratio for the next year to align quarterly extrapolations with that annual BI forecast.
- This paper formalizes the enhanced Denton PFD in matrix form, derives an analytical solution, and compares it to both the QNA Manual shortcut and the basic Denton formula using artificial and real-life examples.

### II. The basic Denton PFD benchmarking method (formulation and properties)
- Problem setup:
  - y: (1)sN× vector of unknown quarterly benchmarked values.
  - y0: (1)N× vector of known aggregated (annual) values.
  - p: (1)sN× vector of preliminary values with Jp ≠ 0.
  - Temporal aggregation constraint: 0 = J y, with J = I_s ⊗ 1′ and s (e.g., s = 4 for quarterly-to-annual).
- Objective (Cholette modified Denton PFD): minimize sum of squared proportional first differences between y and p subject to aggregation constraint. Matrix form: minimize (ŷ − p)′ Q (ŷ − p) s.t. J y = y0 with Q = (D′ diag(p)^{-1} D)^{-1} and D the first-differences matrix.
- Simplified constrained solution provided (formula (4) in paper) can be used for extrapolation by extending J with k zero columns when k additional preliminary high-frequency observations are available.
- Extrapolation property of basic Denton PFD:
  - For extrapolated periods, BI ratio corresponds to the BI ratio of the fourth quarter covered by the last benchmark year.
  - Quarterly growth rates of extrapolated y will match quarterly growth rates of p, implying the annual growth rate of y for year N+1 is implicitly derived from the last available BI ratio and user has no control.

### III. The enhanced Denton PFD method for extrapolation (analytical solution)
- Additional constraint introduced to impose a user forecast of the annual BI ratio for year N+1:
  - Constraint (5): weighted sum of quarterly benchmarked values for year N+1 equals forecast BI ratio times aggregated preliminary series for year N+1.
  - BI forecast parameterization: b_{N+1} = q b_N where q is expected forecast change (expression (8)).
- Matrix expression of added constraint given (expression (9)) and combined constraint system in (10); both distributed and extrapolated values obtained by using formula (4) with modified J* and y0*.
- Advantages of the analytical enhanced solution:
  - Explicitly incorporates user forecast of annual BI ratio for next year.
  - Offers control over annual extrapolated BI ratio and reduces potential bias in purely mechanical Denton extrapolation.
  - Analytical solution differs quarter-by-quarter from both basic Denton and the QNA Manual shortcut; solves the enhanced constrained optimization exactly.

### III.A. QNA Manual shortcut approximation of the enhanced method
- Shortcut constructs adjustment constant η defined by η = (q b_{N+1} − q_{sN−1}) / 3 (expression (11) in the paper) to adjust quarterly BI ratios beginning from second quarter of last available year.
- Recursion to compute extrapolated quarterly BI ratios using η provided.
- Shortcut guarantees the implicit annual extrapolated BI ratio equals the user-requested annual BI ratio, but provides suboptimal quarter-to-quarter movements relative to the analytical solution.

### IV. Artificial-data example (QNA Manual example 6.2 reproduced)
- Indicator series and annual benchmarks for 1998–2000 used to compare:
  - Basic Denton PFD extrapolation (uses last available quarterly BI ratio 10.355 from 1999:Q4).
  - Enhanced Denton PFD with user forecast: assumed BI ratio in 2000 increases 2.0 percent from previous year (resulting BI ratio 10.486 = 10.280 × 1.02).
  - Shortcut enhancement uses η = (10.355 − 10.486) / 3 = −0.044 in example.
- Key numeric outcomes from the example:
  - Indicator annual growth for 2000: 0.9 percent.
  - Basic Denton PFD annual rate for 2000: 1.6 percent.
  - Enhanced Denton PFD (shortcut) annual rate for 2000: 2.9 percent (difference from indicator growth exactly equal to 2.0 percent forecast change in BI ratio).
  - Analytical enhanced solution yields the same annual rate for 2000 as the shortcut (2.9 percent) but different quarter-to-quarter values and growth rates.
- Mean squared differences (MSD) of quarterly growth rates (square root of average of squared differences between growth rates of indicator series and benchmarked series):
  - MSD total (entire sample): Basic Denton PFD 0.5946, Enhanced PFD shortcut 0.6523, Enhanced PFD analytical 0.6392.
  - MSD year 2000 (extrapolated quarters 2000:Q1–2000:Q4): Basic Denton PFD 0.0000, Enhanced PFD shortcut 0.4184, Enhanced PFD analytical 0.3312.
  - Interpretation: basic Denton preserves indicator quarterly movements perfectly in extrapolation (hence MSD 0.0000 for extrapolated year), analytical enhanced is closer to indicator movements than the shortcut in extrapolation, but both enhanced methods achieve the requested annual BI ratio for 2000 (2.9 percent annual rate).
- Wider sensitivity (Table 5): varying b_{2000} over [0.94, 1.06] shows:
  - Analytical enhanced method always closer to indicator movements (lower PFD criterion and MSD) than the shortcut.
  - Improvement of analytical over shortcut is larger as BI ratio change deviates more from 1.0.
  - Basic Denton PFD preserves indicator movements best at extrapolation time but does not allow user control over annual BI ratio.

### V. Application to real-life data: forecasting manufacturing VA for 2009 using IPI
- Experiment objective: derive out-of-sample extrapolation of manufacturing value added (VA) for 2009 using quarterly industrial production index (IPI) of manufacturing as the indicator series via Denton PFD methods.
- Rationale: IPI is a timely indicator often used as a short-term proxy for VA (single-deflation approach); 2008–2009 years saw a strong downturn breaking many historical short-term relationships, including IPI–VA.
- Data and sample:
  - Countries: Australia, Canada, France, Germany, Indonesia, Italy, Japan, Korea, Mexico, South Africa, Turkey and the United States (twelve G-20 countries).
  - Sources: time series extracted from “National Accounts” and “Industry and Services” themes in the OECD data warehouse (http://stats.oecd.org/).
  - Sample period: 2000 through 2009.
  - Extraction date: March 2012 (national accounts data for these years should be consolidated as they incorporate fully revised short-term statistics and annual survey data).
- Expectation: enhanced Denton PFD may improve extrapolation for 2009 relative to basic Denton PFD if the break in the IPI–VA relationship is properly addressed by incorporating a forecasted annual BI ratio.

### VI. Comparative findings and methodological implications
- Basic Denton PFD:
  - Preserves quarterly movements of the indicator in extrapolation (no quarter-to-quarter adjustment), which yields zero MSD for extrapolated year but offers no user control of annual BI ratio.
  - Extrapolation is purely mechanical and may miss systematic bias between indicator and target.
- Enhanced Denton PFD:
  - Allows explicit imposition of a forecast for annual BI ratio b_{N+1} (e.g., via multiplicative factor q relative to b_N).
  - Analytical solution (matrix formulation) provides exact constrained optimization incorporating the BI forecast.
  - Analytical enhanced solution outperforms the QNA Manual shortcut in terms of closeness to indicator movements (lower PFD criterion and MSD), particularly when the requested BI ratio change is larger (further from 1.0).
  - Shortcut guarantees the requested annual BI ratio but yields suboptimal quarter-to-quarter preservation relative to analytical enhanced solution.
- Practical implication for national accounts:
  - Use of enhanced Denton PFD with an analytically derived solution gives users control over anticipated annual BI ratio while preserving quarterly continuity and better matching indicator movements than the shortcut.
  - When the annual benchmark for year N+1 becomes available and matches the BI ratio forecast used in extrapolation, enhanced Denton PFD will reduce future revisions of quarterly data.

*Source: IMF Working Paper — section "6. Forecasting Manufacturing Value Added in 2009 Using IPI: a Comparison Between PFD and EPFD" (text and tables as provided).*

### Section II, the last available quarterly BI ratio is used to extrapolate any subsequent quarters

### Section II, the last available quarterly BI ratio is used to extrapolate any subsequent quarters

### Methodology and problem statement
- With the basic Denton PFD method, the last available quarterly BI ratio is carried forward to extrapolate subsequent quarters not covered by annual benchmarks. This implies:
  - Quarterly benchmarked estimates will show the same quarterly growth rates of the indicator series used.
  - The annual growth rate may differ from that shown by the indicator series.
- The enhanced Denton formula (as in the QNA Manual) requires an input from the user: the forecast of the annual BI ratio for the next year (see equation (7) in source).
- Out-of-sample exercise: assume the BI ratio for 2009 is unknown; derive a forecast for 2009 given BI ratios up to and including 2008.

### BI ratio behavior and forecasting rules
- Observed patterns (2000–2008) justify simple forecasting rules rather than sophisticated time-series models given the low number of observations.
- Interpretation of BI ratio movements:
  - An increase of the BI ratio over time signals that VA is growing faster than IPI.
  - A decrease signals VA growing slower than IPI.
  - A stable BI ratio around a mean denotes a good approximation of VA growth by IPI.
- Forecasting rules used to derive the annual 2009 BI forecast (using 2000–2008 data):
  - When the BI ratio does not show a clear trend and fluctuates around a long-term average, use the average level of the BI ratio observed in the last five years (2004–2008). This approach is used for: Australia, Italy, Mexico.
  - When the BI ratio displays a clear trend movement in the latest years, use the geometric average of the growth rates of the BI ratio observed in the last five years (2004–2008). This approach is used for: Canada, France, Germany, Indonesia, Japan, Korea, South Africa, Turkey, United States.
- Rationale:
  - The five-year span is a subjective choice aiming for a simple, pragmatic forecasting method to improve extrapolation quality relative to the basic Denton benchmarking method.

### Extrapolation experiment and comparative results (Table 6 summary)
- First two columns of Table 6: IPI 2009 % growth and Y GDP (VA) 2009 % growth (the latter mostly negative due to the global financial crisis).
- In most countries IPI growth was more negative than VA growth (exceptions: Germany, Korea, Mexico).
- Basic Denton PFD tends to produce 2009 extrapolated VA growth rates similar to the IPI growth rate but not equal; it does not incorporate historical BI ratio movements.
- Enhanced Denton PFD incorporates the user-provided annual BI forecast and uses historical BI movements to improve extrapolation.

- Country-level results (columns interpreted as: IPI 2009 % growth ; Y GDP 2009 % growth ; Basic Denton PFD Q GDP 2009 % growth ; Basic Denton 2009 error ; Enhanced Denton PFD Q GDP 2009 % growth ; Enhanced Denton 2009 error):

  - Australia*: IPI -7.3 ; Y GDP 1.5 ; Basic Denton Q GDP -9.4 ; Basic error -10.9 ; Enhanced Q GDP -2.3 ; Enhanced error -3.8
  - Canada: IPI -13.9 ; Y GDP -13.6 ; Basic Denton Q GDP -13.9 ; Basic error -0.3 ; Enhanced Q GDP -13.8 ; Enhanced error -0.2
  - France: IPI -13.9 ; Y GDP -11.0 ; Basic Denton Q GDP -13.7 ; Basic error -2.7 ; Enhanced Q GDP -13.5 ; Enhanced error -2.5
  - Germany: IPI -17.0 ; Y GDP -18.1 ; Basic Denton Q GDP -17.8 ; Basic error 0.3 ; Enhanced Q GDP -18.2 ; Enhanced error -0.1
  - Indonesia: IPI 1.4 ; Y GDP 2.2 ; Basic Denton Q GDP 1.8 ; Basic error -0.4 ; Enhanced Q GDP 3.8 ; Enhanced error 1.6
  - Italy*: IPI -19.3 ; Y GDP -16.0 ; Basic Denton Q GDP -19.5 ; Basic error -3.5 ; Enhanced Q GDP -18.7 ; Enhanced error -2.7
  - Japan: IPI -21.9 ; Y GDP -19.7 ; Basic Denton Q GDP -21.4 ; Basic error -1.7 ; Enhanced Q GDP -20.6 ; Enhanced error -0.9
  - Korea: IPI -0.9 ; Y GDP -1.5 ; Basic Denton Q GDP -1.0 ; Basic error 0.5 ; Enhanced Q GDP -1.3 ; Enhanced error 0.2
  - Mexico*: IPI -9.5 ; Y GDP -9.9 ; Basic Denton Q GDP -9.5 ; Basic error 0.4 ; Enhanced Q GDP -9.9 ; Enhanced error 0.0
  - South Africa*: IPI -12.6 ; Y GDP -10.4 ; Basic Denton Q GDP -12.2 ; Basic error -1.8 ; Enhanced Q GDP -11.2 ; Enhanced error -0.8
  - Turkey: IPI -11.5 ; Y GDP -7.2 ; Basic Denton Q GDP -11.1 ; Basic error -3.9 ; Enhanced Q GDP -10.1 ; Enhanced error -2.9
  - United States: IPI -13.5 ; Y GDP -9.7 ; Basic Denton Q GDP -13.8 ; Basic error -4.2 ; Enhanced Q GDP -12.5 ; Enhanced error -2.8

- Aggregate error statistics:
  - Mean error (Basic Denton PFD): -2.3 percent
  - Mean error (Enhanced Denton PFD): -1.2 percent
  - Mean absolute error (Basic Denton PFD): 2.5
  - Mean absolute error (Enhanced Denton PFD): 1.6

- Notes:
  - *For Australia, Italy, Mexico, the annual BI forecast required by the enhanced method is the average level of the BI ratios in 2004–2008.
  - For other countries, the annual BI forecast is the geometric average of the annual growth rates of the BI ratios in 2004–2008.

### Findings and interpretation
- Taking into account historical BI ratio movements improves extrapolation accuracy in most cases.
- Enhanced Denton formula reduced extrapolation errors in absolute terms for all countries considered except Indonesia:
  - Indonesia: enhanced method over-estimated by 1.6 percent vs basic method under-estimated by 0.4 percent.
- Example: Australia shows substantial improvement with the enhanced method:
  - Basic error -10.9 percent vs enhanced error -3.8 percent, reflecting a rebound in the BI ratio in 2009 after the drop in 2008 that the basic method (carrying-forward 2008:Q4) failed to capture.
- Overall average performance improved:
  - Basic Denton: mean error -2.3 percent, mean absolute error 2.5.
  - Enhanced Denton: mean error -1.2 percent, mean absolute error 1.6.

### Conclusions and recommendations
- Limitation of basic Denton PFD:
  - Implicit annual forecast for the next year is effectively determined by the level of the last observed annual BI ratio (and possibly the preceding one), which may be biased by preliminary information and unrelated to the historical relationship between the indicator series and annual benchmarks.
- Advantage of enhanced Denton PFD:
  - Allows explicit introduction of a forecast for the next year’s annual BI ratio, enabling better control of the extrapolation mechanism.
- Practical recommendations:
  - Prefer the analytical matrix solution of the enhanced Denton PFD method presented in this paper rather than the QNA Manual’s shortcut recursion, because the shortcut may distort quarter-to-quarter movements in the indicator series.
  - When a systematic pattern arises from the annual BI ratio series—especially in the latest years—users should exploit this information by applying the enhanced Denton PFD method.
- Caveat:
  - The effectiveness of the enhanced method depends on the quality of the annual BI forecast; poor forecasts (e.g., Indonesia in this exercise) can make the enhanced method perform worse than the basic method.

*Source: _wp12169 - Section II, the last available quarterly BI ratio is used to extrapolate any subsequent quarters*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12169.pdf_