|
on Forecasting |
| By: | Michal Franta |
| Abstract: | This paper examines the effect of non-linearities on density forecasting. It focuses on the relationship between credit markets and the rest of the economy. The possible non-linearity of this relationship is captured by a threshold vector autoregressive model estimated on the US data using Bayesian methods. Density forecasts thus account for the uncertainty in all model parameters and possible future regime changes. It is shown that considering non-linearity can improve the probabilistic assessment of the economic outlook. Moreover, three illustrative examples are discussed to shed some light on the possible practical applicability of density forecasts derived from non-linear models. |
| Keywords: | density forecasting, nonlinearity, threshold autoregressive model. |
| JEL: | C11 C32 E44 |
| Date: | 2013–09 |
| URL: | http://d.repec.org/n?u=RePEc:cnb:wpaper:2013/09&r=for |
| By: | Hull, Isaiah (Research Department, Central Bank of Sweden) |
| Abstract: | I construct an estimable statistic that predicts whether a financial innovation will spread. The approach embeds the multi-host SIR model from epidemiology within a financial model of correlated securities trade; and takes advantage of the related predictive tools from mathematical epidemiology, including the basic reproductive ratio (R0) and herd immunity. In the model, banks and their creditors are assumed to have imperfect information about a newly-created security, and must search over the portfolios of other investors and intermediaries to infer the security's properties. In the absence of historical returns data, a large mass of firms holding the new security and not experiencing insolvency provides a positive signal about the distribution of its returns within the current period, and perpetuates further holding of the security. The model yields a set of structural equations that are used to construct the statistic. I provide two estimation strategies for the statistic; and identify 12 theoretical parameter restrictions that enable inference when only a subset of the model's parameters are identifiable. I use the approach to predict the spread of exchange traded funds (ETFs) and asset-backed securities (ABS). Additionally, I show how regulators can use the method to monitor the joint solvency of depository institutions within a given geographic region. |
| Keywords: | Econometric Modeling; Econometric Forecasting; Financial Econometrics; Financial Innovation |
| JEL: | C51 C53 G12 G14 |
| Date: | 2013–10–01 |
| URL: | http://d.repec.org/n?u=RePEc:hhs:rbnkwp:0279&r=for |
| By: | Ivan Kitov; Oleg Kitov |
| Abstract: | We re-estimate statistical properties and predictive power of a set of Phillips curves, which are expressed as linear and lagged relationships between the rates of inflation, unemployment, and change in labour force. For France, several relationships were estimated eight years ago. The change rate of labour force was used as a driving force of inflation and unemployment within the Phillips curve framework. The set of nested models starts with a simplistic version without autoregressive terms and one lagged term of explanatory variable. The lag is determined empirically together with all coefficients. The model is estimated using the Boundary Element Method (BEM) with the least squares method applied to the integral solutions of the differential equations. All models include one structural break might be associated with revisions to definitions and measurement procedures in the 1980s and 1990s as well as with the change in monetary policy in 1994-1995. For the GDP deflator, our original model provided a root mean squared forecast error (RMSFE) of 1.0% per year at a four-year horizon for the period between 1971 and 2004. The rate of CPI inflation is predicted with RMSFE=1.5% per year. For the naive (no change) forecast, RMSFE at the same time horizon is 2.95% and 3.3% per year, respectively. Our model outperforms the naive one by a factor of 2 to 3. The relationships for inflation were successfully tested for cointegration. We have formally estimated several vector error correction (VEC) models for two measures of inflation. At a four year horizon, the estimated VECMs provide significant statistical improvements on the results obtained by the BEM: RMSFE=0.8% per year for the GDP deflator and ~1.2% per year for CPI. For a two year horizon, the VECMs improve RMSFEs by a factor of 2, with the smallest RMSFE=0.5% per year for the GDP deflator. |
| Date: | 2013–11 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:1311.1097&r=for |