外国為替・株等の金融市場は極めて効率的であり, 外国為替交換レートや株価・株価指標値等の短期金融時系列は, 外部情報に対する瞬時的な応答を除けば, 独立増分を持つ確率過程で近似できると考えられる。しかし実際には, 当該応答の影響は暫時残り, 増分は事前の動作・状態に依存している。本研究では, 事前状態を, トレンドという人間の直観に合う形での記述を試み, また非線形関数で暗黙的に記述し, 予測精度の向上を目指した。前者ではランダム性が強く出, 予測精度の改善はなかったが, 後者では十分な改善効果が得られた。
Financial markets where foreign currency and stocks are exchanged are very efficient and short term time series of values of exchange rates, prices, and indexes are well approximated by stochastic processes with independent increments except for sudden changes caused by outer world events. But in reality, the changes remain and the increments are dependent on previous states/movements. In this research, so-called trends as human experts recognize are tried, and implicit non-linear representations are also tried, and used to describe and predict the time series. We showed that the former results in strong random sequences and does not contribute to improvements of prediction but the latter does.
|
