Scholarlink

<p>Exchange rate forecasting is an important and well-studied problem in finance. However, most existing approaches rely only on historical price data and overlook the role of public sentiment, which can be a strong signal during periods of political and economic instability. This paper presents a hybrid forecasting framework for the USD/PKR currency pair that combines historical exchange rate time series with daily sentiment scores extracted from Urdu-language tweets. From January 2021 to January 2025, a total of 172,002 tweets were gathered from the X (formerly Twitter) platform with the help of 1,079 trending hashtags. As a result of pre-processing and language filtering, 45,048 tweets containing Urdu language were left for sentiment analysis and are contextually relevant. Four methods of sentiment annotation were implemented and compared: Gemini 1.5 Flash, a fine-tuned version of GPT-3.5 Turbo, GPT-4o, and XGBoost trained on FastText embeddings. The SI generated by each model was then matched with the exchange rate data of USD/PKR from the State Bank of Pakistan and fed to three models i.e. Long Short-Term Memory (LSTM), Xtreme Gradient Boosting (XGBoost) and two-stage hybrid LSTM+XGBoost. The hybrid model with GPT-4o based sentiment performed the best with Root Mean Squared Error (RMSE) of 0.0831 and Mean Absolute Percentage Error (MAPE) of 0.03%. The results are better than that of the LSTM baseline trained with historical data alone and similar studies related to the USD/PKR forecasting. The findings show that public opinion in the Urdu social media can be a valuable tool to predict the evolution of exchange rates and that hybrid architectures are more appropriate than standalone models to exploit public opinion in Urdu social media for the purpose of predicting the exchange rates.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting USD PKR exchange rates using Urdu social media sentiment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月10日

查看全文
  • Abid Sohail,
  • Muhammad Rizwan,
  • Muhammad Salman Ali,
  • Javed Rashid,
  • Sultan Hussain,
  • Hezam Gawbah

摘要

Exchange rate forecasting is an important and well-studied problem in finance. However, most existing approaches rely only on historical price data and overlook the role of public sentiment, which can be a strong signal during periods of political and economic instability. This paper presents a hybrid forecasting framework for the USD/PKR currency pair that combines historical exchange rate time series with daily sentiment scores extracted from Urdu-language tweets. From January 2021 to January 2025, a total of 172,002 tweets were gathered from the X (formerly Twitter) platform with the help of 1,079 trending hashtags. As a result of pre-processing and language filtering, 45,048 tweets containing Urdu language were left for sentiment analysis and are contextually relevant. Four methods of sentiment annotation were implemented and compared: Gemini 1.5 Flash, a fine-tuned version of GPT-3.5 Turbo, GPT-4o, and XGBoost trained on FastText embeddings. The SI generated by each model was then matched with the exchange rate data of USD/PKR from the State Bank of Pakistan and fed to three models i.e. Long Short-Term Memory (LSTM), Xtreme Gradient Boosting (XGBoost) and two-stage hybrid LSTM+XGBoost. The hybrid model with GPT-4o based sentiment performed the best with Root Mean Squared Error (RMSE) of 0.0831 and Mean Absolute Percentage Error (MAPE) of 0.03%. The results are better than that of the LSTM baseline trained with historical data alone and similar studies related to the USD/PKR forecasting. The findings show that public opinion in the Urdu social media can be a valuable tool to predict the evolution of exchange rates and that hybrid architectures are more appropriate than standalone models to exploit public opinion in Urdu social media for the purpose of predicting the exchange rates.